Ebook Icesit2016 Final PDF

ICESIT 2016
The 2016 International Conference on Embedded Systems and Intelligent Technology
August 5-8, 2016
Smart City
th
Faculty of Engineering, 8 oor
Burapha University, Thailand
Table of
Contents
ICESIT 2016
The 2016 International Conference on Embedded Systems

and Intelligent Technology
Page
Program Overview..1
Patrons.4
List of Committee Members..5
Welcome Messages...6
Keynote Speakers..7
Invited Talks15
Industrial Session..26
Technical Program.37
Map..119
Program Overview
August 5, 2016
August 6, 2016
Room M801
Session 1: Invited Talks #1-9,
09:00-12:00
Room M801
Session 3: Keynote #1-3,

Invited Talks #10
Coee Break
14:30-14:40
Room M801
Session 4: Industrial Session

14:40-17:30
Room M800
Session 2: Oral Presentation of

Ubiquitous Computing and
Intelligent Systems (UCI)
Lunch
12:00-13:00
13:00-14:30
ICESIT 2016

Registration
17:30-18:00
Room M800
Session 5: Oral Presentation of

Embedded Systems, Electric
Vehicle, Energy Technology, and
Smart Grid (EEES)
Going to Dinner
Session 2: Paper #6, 7, 8, 10, 12, 13, 16, 21, 24

Session 5: Paper #9, 11, 17, 18, 19, 20, 22, 23, 25
Session 1: Room 1 (M801)

Time Presentation
09:00-09:20 Invited Talk 1: Wimol (The Synchronization of Back-to-Back Twisted Chaotic
Oscillators for Secured RF Transmission)
09:20-09:40 Invited Talk 2: Wimol (Reviews on Evidence-Based Pulsed Electromagnetic
Field (OEMF) Therapy in Medical Clinical Surgery)
09:40-10:00 Invited Talk 3: Prabhas (Machine Learning technique to improve performance
of cache bypassing)
10:00-10:20 Invited Talk 4: Zadeh (Security in Embedded Systems)
10:20-10:40 Invited Talk 5: Kitsana (Evolution-based method for clustering of high
dimensional data streams)
10:40-11:00 Invited Talk 6: Suppakit and Chuwong (Optimization Algorithm Integrated on
Adaptive Array Antenna for Emerging Wireless Communication System)
11:00-11:20 Invited Talk 7: Wirot (Eye Gaze Tracking System)
11:20-11:40 Invited Talk 8: Sorawat (A New Design of Variable Fractional-Delay FIR Filter
and Its Low-Complexity Filter Structure)
11:40-12:00 Invited Talk 9: Navapadol (The Electric Field Distribution of a Hybrid
Rectangular and Circular Waveguide Resonator for Use in Granular Material
Dielectric Measurements)
ICESIT2016 Page 1 of 119
Program Overview
ICESIT 2016


Time Presentation
13:00-13:20 Keynote 1: Industry 4.0 (Watchara)
13:20-13:40 Keynote 2: IoT Smart Cities (Wiroon)
13:40-14:00 Keynote 3: Open Platform for IoT & Smart City Data Exchange (Akkarit)
14:00-14:20 Invited Talk 10: Smart EV: Charging, Driving and Connecting (Werachet)

Time Presentation
09:00-09:20 UCI: Paper 6 (Su-Min Hong and Yo-Sung Ho. Virtual View Synthesis Using
Moving Multi-Camera Array)
09:20-09:40 UCI: Paper 7 (Chontawee Wongpromrat and Watchara Chatwiriya. Relayed
Message Sending for MQTT Protocol)
09:40-10:00 UCI: Paper 8 (Wisarut Chantara and Yo-Sung Ho. Image Fusion using Image
Blocks and Modified Discrete Wavelet Transform)
10:00-10:20 UCI: Paper 10 (Jaeryun Ko and Yo-Sung Ho. Census Transform based Stereo
Matching with Gradient Stereo Images)
10:20-10:40 UCI: Paper 12 (Prasit Nantin, Pinit Kumhom and Kosin Chamnongthai.
Adaptive Silhouette Normalization by Traditional of the Dierential of Neighboring
SilhouetteI)
10:40-11:00 UCI: Paper 13 (Pattanai Sirisombatampai and Werapon Chiracharit. Automatic
Comparison System for IC Units Positions)
11:00-11:20 UCI: Paper 16 (Apirath Limmanee, Sorakrai Kraipui, Wiroon Sriborrirux, Nipon
Sommai and Soontree Jaikhong. Adaptive Hybrid Encryption in Resource-andBandwidth-Constrained Systems with Secret Mixing)
11:20-11:40 UCI: Paper 21 (Montri Phothisonothai, Suchada Tantisatirapong and Apinan
Aurasopon. Watermelon Ripeness Extraction Based on Image Color
Segmentation and Rind Texture Analysis)
11:40-12:00 UCI: Paper 24 (Choawanat Samkampang and Daranee Hormdee. LowComplexity Lossless Image Compression for Low-Performance Microcontroller
based on Color-Region Segment)
Program Overview
ICESIT 2016


Time Presentation
14:40-14:55 EEES: Paper 9 (Supachai Vorapojpisut. Educating Model-based Development
in the 9th TESA Top Gun Rally)
14:55-15:10 EEES: Paper 11 (Tanapong Piansompon and Supachai Vorapojpisut. A Realtime Extension of Simulink Support Packages for Arduino)
15:10-15:25 EEES: Paper 17 (Anuparp Boonsongsrikul, Nakrom Massaya and Jirayut
Phangklin. Circuit Analysis for Heartbeat Measurement)
15:25-15:40 EEES: Paper 18 (Werachet Khan-Ngern. Design of Zero Energy Consumption
Home Using Stand Alone Solar System)
15:40-15:55 EEES: Paper 19 (Werachet Khan-Ngern. Design of Automated Guided Vehicle
for Industry)
15:55-16:10 EEES: Paper 20 (Tossaporn Wetsiri and Wimol San-Um. The Implementation
of a Wireless-Remote Arduino- Compatible Programmable Logic Controller with
Smart User Interface towards Industrial Internet of things)
16:10-16:25 EEES: Paper 22 (Suttichai Premrudeepreechacharn, Kanchit Ngamsanroaj and
Khambai Vongsomphone. Optimal Load Dispatch of Hydro Power Plant in
Cental-1 of Laos Using Artificial Neural Network)
16:25-16:40 EEES: Paper 23 (Suttichai Premrudeepreechacharn and Kanchit Ngamsanroaj.
Reliability Assessment of the Electrical Distribution System of 22 kV of the
Khuatinueng Substation in Luangprabang Lao PDR)
16:40-16:55 EEES: Paper 25 (Surin Sermsukroongsakul and Suttichai
Premrudeepreechacharn. Remaining Lifetime Study of Generator Step-Up
Transformers in Thailand)
ICESIT 2016

Patrons
List of
Committee Members
ICESIT 2016

Advisory Board
Honorary Chairs
1. Sompol Pongthai (BUU, Thailand)
1. Prabhas Chongstitvatana (CU, Thailand)
2. Anat Deepatana (BUU, Thailand)
2. Kosin Chamnongthai (KMUTT, Thailand)
3. Apinetr Unakul (TESA, CAT Telecom)
General Co-chairs
4. Pansak Siriruchatapong (MICT)
1. Wiroon Sriborrirux (BUU, Thailand)
5. Narong Yoothanom (SPU, Thailand)
2. Takao Onoye (Osaka U, Japan)
6. Werachet Khan-Ngern (TESA & KMITL, Thailand)

7. Kang Hyeon RHEE (Chosun, South Korea)
3. Yo-Sung Ho (GIST, Korea)
8. Yoshikazu Miyanaga (Hokkaido, Japan)
General Chair
Panuwat Danklang (BUU, Thailand)
Technical Program Chairs

1. Apirath Limmanee (BUU, Thailand)
19. Daranee Hormdee (KKU, Thailand)
2. Sanchai Eardprab (BUU, Thailand)
20. Watchara Chatwiriya (KMITL, Thailand)
3. Somjet Suppharangsan (BUU, Thailand)
21. Akkarit Sangpetch (KMITL, Thailand)
4. Anuparp Boonsongsrikul (BUU, Thailand)
22. Thumrongrat Amornraksa (KMUTT, Thailand)
5. Goparanjan Mohapatra (BUU, Thailand)
23. Wimol San-um (TNI, Thailand)
6. Norrarat Wattanamongkhol (BUU, Thailand)
24. Karn Opasjumruskit (Silicon Craft, Thailand)
7. Jantana Panyavaraporn (BUU, Thailand)
25. Nattha Jindapetch (PSU, Thailand)
8. Nayot Kurukitkoson (BUU, Thailand)
26. Supaporn Kiattisin (MU, Thailand)
9. Chalermpan Fongsamut (BUU, Thailand)
27. Preecha Kocharoen (SPU, Thailand)
10.Tharathorn boonsri (BUU, Thailand)
28. Sutthichai Premruedeepreechacharn
11. Jedsada Saai (BUU, Thailand)
(CMU, Thailand)
12. Surachat Lekkham (BUU, Thailand)
29. Wimol San-Um (TNI, Thailand)
13. Supachai Vorapojpisut (TU, Thailand)
30. Virasit Imtawil (KKU, Thailand)
14. Wannarat Suntiamorntut (PSU, Thailand)
31. Somyot Wattanakamolchai (KMITL, Thailand)
15. Pinit Kumhom (KMUTT, Thailand)
32. Lunchakorn Wuttisittikulk (CU, Thailand)
16. Vara Varavithya (KMUTNB, Thailand)
33. Tharadol Komolmis (CMU, Thailand)
17. YodYium Tipsuwan (KU, Thailand
34. Soradech Krootjohn (KMUTNB, Thailand)
18. Akaraphunt Vongkunghae (NU, Thailand)
Message from
TESA President
ICESIT 2016

Welcome all speakers, participants and exhibitors to the 13th International

Conference on Embedded System to Intelligent Technology, ICESIT 2016. This
year, 13th ICESIT committee proud to announce the theme of SMART CITY to
promote the growth of IOT application. Many thanks to all members of Department
of Electrical Engineering, Burapa University to host the 2016 ICESIT at Bangsaen
city. Bangsaen city is a lovely atmosphere for tourists and very interesting place to
visit.
I wish the 2016 ICESIT meet the goal to be the forum among academic,
researchers and industrial section to learn, share and exchange information on IOT.
The network of ICESIT members is enhanced and expanded to not only industry,
but also to social activities.
Please enjoy your visit and your stay in Bangsaen, Thailand.

Werachet Khan-ngern
Thai Embedded System Association, President
ICESIT 2016
August 5-8, 2016
Keynote Speakers
Keynote Speaker
Watchara Chatwiriya
ICESIT 2016

Dr.Watchara Chatwiriya
Lecturer of Computer Engineering Dep.,
Faculty of Engineering,
King Mongkut's Institute of Technology Ladkrabang
Email: watchara.ch@kmitl.ac.th
Education
2002
Ph.D in Computer Engineering
West Virginia University, USA
1992
Master of Computer Engineering
King Mongkut's Institute of

Technology Ladkrabang
1985
Bachelor of Computer Engineering
King Mongkut's Institute of

Technology Ladkrabang
Area of Expertise
Management Information System
Computer Network and communication
ICT Master Plan Development
Teaching Class
Signal Processing & Recognition

Object Oriented Design and Analysis
Software Engineering
Enterprise Architecture
Experience
2015 - Present
Board of Director
Bank for Agricultural and Cooperative Agricultural
2011 - Present
ICT Advisor
Office of The Narcotics Control Board (Bangkok)
2008 - Present
Secretary
Thai Embedded System Association
2005 - 2007
Vice Director
Computer Research & Service Center, KMITL
2003 - 2005
Department Chair
Department of Computer Engineering, KMITL
1978 - Present
Lecturer
Department of Computer Engineering, KMITL
Keynote Speaker
Watchara Chatwiriya
ICESIT 2016

Research and Publication

1. N. Vatanasuk, A. Chomputawat, S. Chomputawat and W. Chatwiriya, Mobile Crime Incident
Reporting System using UX dimensions guideline, 2015 Asian Conference on Defence
Technology (ACDT 2015), 2015
2. W. Puisamlee and W. Chatwiriya, RFID Application in Food Processing, International
Conference on Embedded Systems and Intelligent Technology (ICESIT 2014), Korea, 2014
3. K. Rongsawad and W. Chatwiriya, Expressive Speech Classification, International
Conference on Embedded Systems and Intelligent Technology (ICESIT 2014), Korea, 2014
4. Surit S. and Chatwiriya W., Forest Fire Smoke Detection in Video Based on Static and
Dynamic Characteristic Analysis, International Conference on Embedded Systems and
Intelligent Technology (ICESIT 2012), 2012
5. Surit S. and Chatwiriya W., Forest Fire Smoke Detection in Video Based on Digital Image
Processing Approach with Static and Dynamic Characteristic Analysis, First ACIS/JNU
International Conference on Computers, Networks, Systems and Industrial Engineering (CNSI
2011), 2011
6. Rungrittidetch W. and Chatiwiriya W., Camera Based Document Scanner, International
Conference on Embedded Systems and Intelligent Technology (ICESIT 2011), 2011
7. Suwannawet P. and Chatwiriya W., Reducing training set in Conditional Random Field Thai
Word Segmentation using word frequency, National Conference in Computer and Information
Technology, 2011
8. Ruanthai K. and Chatwiriya W., Stereoscopic Depth Perception Enhancement with
Monocular Depth Cue, Joint International Conference on Information & Communication
Technology Electronic and Electrical Engineering (JICTEE 2010), 2010
9. PKotchakoon P. and Chatwiriya W., Image Enhancement with HDRI for License Plate
Recognition System, Joint International Conference on Information & Communication
Technology Electronic and Electrical Engineering (JICTEE 2010), 2010
10. Rojanavipark S. and Chatwiriya W., Real-Time License Plate Localization on an Embedded
DSP using Edge processing and Adaboost technique (EECON2010), 2010
11. Chatwiriya, W.; Klinkhachorn, P.; Lass, N., Thai Handwriting Legal Amounts Recognition,
Proceedings of the 35th Southeastern Symposium on System Theory, 2003
12. Klinkhachorn, P. Nomani, S. Chatwiriya, W. Halabe, U.B., Development of A Portable
Ultrasonic Timber Properties Monitoring Device, Southeastern Symposium on System Theory,
1998
13. Chalamwong Y, Paitoonpong S., and Chatwiriya W., Establishment of Manpower Demand
and Supply Database to Support the Development of Industrial Sector (II), , 2010
14. Chalamwong Y, Paitoonpong S., and Chatwiriya W., Establishment of Manpower Demand
and Supply Database to Support the Development of Industrial Sector (I), , 2009
15. Chalamwong Y, Paitoonpong S., and Chatwiriya W., Increasing the Eectiveness of Labour
Market Data Collection for Preparing Provincial Manpower Plan (II) , 2009
16. Chalamwong Y, Paitoonpong S., and Chatwiriya W., Increasing the Eectiveness of Labour
Market Data Collection for Preparing Provincial Manpower Plan (I) , 2008
17. Chalamwong Y, and Chatwiriya W., Setting up a Better Management Administration System
to Facilitate Manpower Development Planning of Thai Industrial Sector, 2008
Keynote Speaker
Wiroon Sriborrirux
ICESIT 2016

Associate Professor WIROON SRIBORRIRUX

Lecturer of Electrical Engineering Department,
Faculty of Engineering, Burapha University
Email: wiroon@eng.buu.ac.th
Education
2000 - 2003
M.Sc. Eng. (Computer Science

and Engineering)
Korea University, South Korea
1995 - 1998
Bachelor of Computer
Engineering
Prince of Songkla University, Thailand
Research Area
Embedded System
Wireless Sensor Network and Intelligent

Transportation System (ITS)
Embedded System Development

Embedded OS Development
Real-time OS Development
Remote Cryptography-based RFID Network
Development
Routing Network Protocol Development
RF System-on-Chip Network Stack
Development
Mobile Wireless Network Performance (based
Ns-2 Simulator and ARM9-based Testbed)
Internet of Things
6LowPAN and iOT Gateway
1.
2.
3.
Wiroon Sriborrirux, Panuwat Dan-klang, Wasan Wiyarun, Nakorn Indra-Payoong (2015).

An embedded RF-based motorcycle trajectory data for security monitoring system over
university vehicular network. IEEE 81st Vehicular Technology Conference (VTC2015-Spring),
Glasgow, Scotland, 1-5.
Weradech Kumtong, Panuwat Danklang, Wiroon Sriborrirux (2015). Pin Set Sequence
selection guideline routing for Printed Circuit Board Routing. Knowledge and Smart
Technology (KST), 2015 7th International Conference, 126-130.
Apirath Limmanee, Wiroon Sriborrirux, Sorakrai Kraipui (2015). Hybrid Encryption Scheme
for Digital Content with Key Partitioning and Secret Mixing: Design and Implementation. 10th
International ITG Conference on Systems, Communications and Coding (SCC 2015), 1-6.
Keynote Speaker
Wiroon Sriborrirux
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
ICESIT 2016

Wiroon Sriborrirux, Panuwat Promsiri, and Apirath Limanee (2014). Multiple Secret Key
Sharing based on Network Coding Technique for Open Cloud DRM Service Provider. IEEE
17th International Conference on Computational Science and Engineering, 953-959.
Wiroon Sriborrirux, Pariwat Leamsumran, Panuwat Dan-klang (2014). Real-time system
for monitoring activity among the elderly using an RF SoC device with triaxial accelerometer
data over a wireless sensor network. The 2014 IEEE MTT-S International Microwave
Workshop Series on RF and Wireless Technologies for Biomedical and Healthcare
Applications (2014)
Wiroon Sriborrirux, Apirath Limmanee, and Sorakrai Kraipui (2013). Unequal Security
Protection: A Unified Framework, Implementation, and Performance Evaluation of Theoretical
and Practical Security. The Tenth International Symposium on Wireless Communication
Systems 2013
Khwanchai Sringiwrai, Nakorn Indra-Payoong, Agachai Sumalee, Pairoj Raothanachonkun,
Wiroon Sriborrirux. RFID-based travel time estimation: development case in Bangkok.
Conference Proceedings of the 15th HKSTS International Conference (Transportation and
Urban Sustainability) (2010)
Wiroon Sriborrirux, S. Kraipui, N. Indra-Prayoong (2009). B-VIS Service-oriented
Middleware for RFID Sensor Network. Singapore 2009 International Conference on Sensor
Networks, Information, and Ubiquitous Computing held in Singapore (2009), 836840.
Wiroon Sriborrirux, Danklang, P., and Indra-Payoong, The Design of RFID Sensor
Network for Bus Fleet Monitoring, In: Proceedings of the 8th International Conference on
Intelligent Transport System Communications, 22-24 October, Phuket, Thailand pp. 103
107, 2008
Warathon, Sumonrat, Wiroon Sriborrirux, Applied Cluster Computing for NP-Hard
Problem, Industrial and Research Projects of Thailand Research Fund, Thailand, 2006
WonJun, Wiroon Sriborrirux, Optimizing A Secure Routing Protocol Using ID-Based
Cryptography in Wireless Mobile Ad Hoc Networks, LNCS (Lecture Note of Computer
Science), Springer, USA, 2004
Wiroon Sriborrirux, Optimizing A Secure Routing Protocol Using ID-Based Cryptography
in Wireless Mobile Ad Hoc Networks, International Conference of the ICOIN2004, Busan,
South Korea, 2004
Wiroon Sriborrirux, DIAMETER strong Security Extensions using Kerberos v5 in Wireless
LAN, Proceeding of the KIPS, Seoul, South Korea, 2002
Wiroon Sriborrirux, Implementation of IPSec over MPLS-based MIP supporting AAA
Server, Proceeding of the KIPS, Seoul, South Korea, 2001
Keynote Speaker
ICESIT 2016

Akkarit Sangpetch
Mr. Akkarit Sangpetch, Ph.D.

Lecturer, Department of Computer Engineering
Faculty of Engineering
King Mongkuts Institute of Technology Ladkrabang
1 Soi Chalongkrung 1, Ladkrabang
Bangkok, Thailand 10520
Email: akkarit.sa@kmitl.ac.th, akkarit@live.kmitl.ac.th
Website: http://soup.ce.kmitl.ac.th/faculty/akkarit/
Education
2005 - 2013
Doctor of Philosophy
Electrical and Computer Engineering

Carnegie Mellon University, USA
2005 - 2010
Master of Science

2001 - 2005
Bachelor of Science
with University Honors
Computer Science
Bachelor of Science
with University Honors

Work Experience
2013 - Present
Lecturer, Department of Computer Engineering, Faculty of Engineering

King Mongkuts Institute of Technology Ladkrabang (KMITL), Bangkok,
Thailand
2012 - 2013
Member of Technical Sta Cloud Infrastructure Management VMware,

Inc. Palo Alto, CA, USA
Publications
Akkarit Sangpetch, Hyong S. Kim, VDEP: VM Dependency Discovery in Multi-tier

Cloud Applications, in Proceedings of IEEE 8th International Conference on Cloud
Computing (CLOUD 2015), New York, NY, USA, 2015.
Akkarit Sangpetch, Orathai Sangpetch, Self-Scaling Platform as a Service, in
Proceedings of the International Conference on Engineering Science and Innovative
Technology (ESIT 2014), Krabi, Thailand, 2014.
Akkarit Sangpetch, Tactic: Trac Aware Cloud for Tiered Infrastructure Consolidation,
Ph.D. Dissertation, Carnegie Mellon Univ., Pittsburgh, PA, USA, 2013.
Orathai Sukwong, Akkarit Sangpetch, Hyong S. Kim, SageShift: Managing SLAs for
Highly Consolidated Cloud, in Proceedings of IEEE INFOCOM 2012, Orlando, FL, USA,
2012.
Keynote Speaker
Akkarit Sangpetch
ICESIT 2016

Akkarit Sangpetch, Andrew Turner, Hyong S. Kim, How to Tame Your VMs: An
Automated Control System for Virtualized Services, in Proceedings of the Large
Installations Systems Administration (LISA) Conference, San Jose, CA, USA, 2010.
Andrew Turner, Akkarit Sangpetch, Hyong S. Kim, Empirical Virtual Machine Models for
Performance Guarantees, in Proceedings of the Large Installations Systems
Administration (LISA) Conference, San Jose, CA, USA, 2010.
Research
2015 - 2016
LIFE: Scalable Log Analysis and Intrusion Detection System

with Chanon Jaratsuttikul, Thanaphon Soo, Orathai Sangpetch
Runner-up in Thailands 2016 National Software Contest (NSC 18),
Science and Technology Development Project
Thoth: Dynamic Resource Management for Platform-as-a-Service
with Nut Juangmariskul, Supakorn Warodom, Orathai Sangpetch
Runner-up in Thailands 2016 National Software Contest (NSC 18),
Linux Application Development
Tiramisu: Software-defined Data Platform
with Rinlapat Loetthanakulpong, Nattapong Amornbunchornvej,
Orathai Sangpetch
Honorable mention in Thailands 2016 National Software Content
(NSC 18),
Linux Application Development
2014 - 2015
Automated Security Assessment for Data Center

with Sarun Siriphantnon, Supol Prachachon, Orathai Sangpetch
Cybersecurity War Game
with Sermsak Thammaratanarangsri, Adisorn Kamwhang, Orathai Sangpetch
Graphical Modeling and Interactive Visualization for Data Center Management
with Noppakorn Chaiboonruang, Nontaya Wilailertsombat, Orathai Sangpetch,
Second Runner-up in Thailands 2015 National Software Contest (NSC 17),
Science and Technology Development Project
Distributed Cloud Monitoring and Real-time Analysis
with Peerawit Praphanwittaya, Apirat Puttaraksa, Orathai Sangpetch
2010 - 1013
Distributed Resource Scheduling for SLA-oriented Cloud

with Prof. Hyong S. Kim, Carnegie Mellon University, USA
2009 - 2010
Virtual Resource Management for Multi-tier Applications

2008 - 2009
Service-oriented Virtual Machine Provisioning

2006 - 2007
Visualizing Distributed Firewall in Large-scale Service Providers

with Portugal Telecom, Prof. Hyong S. Kim
2005 - 2006
Argus Network Security Management System

with Carnegie Mellon CyLab
Keynote Speaker
Akkarit Sangpetch
ICESIT 2016

Teaching Experience
B.Eng. in Computer Engineering: Cloud Computing, Operating Systems, Unix Operating
System, Object Oriented Analysis & Design
M.Eng. in Computer Engineering: Special Topics in Software Technology
Guest Speaker
2014-2016: Trainer for Thailands International Olympiads in Informatics (IOI) contestants at
The Institute for the Promotion of Teaching Science and Technology
December 2, 2015: Guest speaker Cloud computing direction for 2016 in Leveraging the
Cloud seminar, held by KSC Commercial Internet Co. Ltd.
August 23, 2013: Guest speaker Cloud Technology and related research in seminar for
M.S. in Computer Science students, Kasetsart University
Academic Services
2014-2016: IOI Contestant selection subcommittee member, Institute for the Promotion of
Teaching Science and Technology, Ministry of Education
2014-2016: Information technology project screening subcommittee member, King
Mongkuts Institute of Technology Ladkrabang
2015-2016: Computer innovation engineering curriculum development committee member,
Faculty of Engineering, King Mongkuts Institute of Technology Ladkrabang
2014-2015: Python programming language textbook for public schools editorial board
member, Institute for the Promotion of Teaching Science and Technology, Ministry of
Education
2014-2015: Computer engineering graduate program management committee member,
Faculty of Engineering, King Mongkuts Institute of Technology Ladkrabang
2013-Present: Information Technology Advisor, Netbay Co. Ltd.
2013-Present: Cloud Technical Advisor, Internet Thailand Public Co. Ltd.
ICESIT 2016
August 5-8, 2016
Invited Talks
The Synchronization of Back-to-Back Twisted Chaotic

Oscillators for Sucured RF Transmission
Wimol San-Um
Intelligent Electronic System Research Laboratory
Faculty of Engineering, Thai-Nichi Institute of Technology
Pattanakarn 37, Saunlaung, Bangkok, Thailand, 10250
Tel: (+662) 763-2600 Ext. 2926, Fax: 0-2763-2700, E-mail: wimol@tni.ac.th.
Abstract
A new simple chaotic system with both dynamic and jerk forms is presented through the use
of an inherent hyperbolic sine function in anti-parallel diode configuration. The circuit
reveals a complex back-to-back twisted chaotic attractor which has never been found in other
circuit configuration. Dynamic properties are described through an existence of attractor,
Equilibria, Jacobian matrices, bifurcations, and time-domain chaotic waveforms. The
proposed chaotic oscillator is applied for secured RF transmission through synchronization
between transmitter and receiver.
Keywords; Chaotic oscillator, RF Transmission, Bifurcation.
Reviews on Evidence-Based Pulsed Electromagnetic Field

(OEMF) Therapy in Medical Clinical Surgery
Wimol San-Um
Intelligent Electronic System Research Laboratory
Faculty of Engineering, Thai-Nichi Institute of Technology
Pattanakarn 37, Saunlaung, Bangkok, Thailand, 10250
Tel: (+662) 763-2600 Ext. 2926, Fax: 0-2763-2700, E-mail: wimol@tni.ac.th.
Abstract
Pulsed Electromagnetic Field (PEMF) technologies have been useful in terms of adjunctive
therapy for the treatment of both delayed-union fractures and chronic wounds. Such a PEMF
is a relatively simple device, which employs an external, non-invasive PEMF to generate
shorts bursts of electrical current in injured tissue without producing heat or interfering with
nerve or muscle function. This paper aims to review major scientific breakthroughs and
current understanding of the mechanism of action of PEMF therapy. A literature review is
conducted, involving mechanism of action and biologic and clinical studies of PEMF. This
review shows that surgeons have a powerful tool with no known side effects for the
adjunctive, noninvasive, nonpharmacologic management of postoperative pain and edema.
Recent rapid advancement of portable and economical PEMF devices have revealed that the
PEMF devices are significant to the surgeon for improving outcomes of decreased pain and
swelling following injury or surgery.
Keywords; Pulsed Electromagnetic Field, therapy, swelling following injury, surgery.
Machine Learning technique to improve performance of cache bypassing

Warisa Sritriratanarak and Prabhas Chongstitvatana
Department of computer engineering, Chulalongkorn University
prabhas.c@chula.ac.th
Abstract
As the processor-memory performance gap increased, the multilevel cache is introduced. For
example, the Intel Pentium processor has level 1 on-chip cache and level 2 off-chip cache
implemented; the L2 cache moved up to be on-chip in the Intel Pentium Pro. With the advent of
multicore processors, several private and shared levels of cache are used, i.e., the Intel
Core Xeon X5550 has four cores and three layers of cache: each core has private L1 and L2
cache; the Last-level cache (LLC) is shared among four cores. Sharing the LLC can reduce the
number of duplicate copies when many cores are running identical applications.
This work concerns with non-inclusion cache which a datum can have multiple copies like an
inclusive cache but when a datum is discarded from the lower level cache, the copies in the upper
levels are not required to be discarded too. When the datum moves up from lower level to higher
level, the exclusive cache will force deletion of the copy in lower level while the non-inclusive
cache will allow leaving the copy in lower level, avoid wasting any blocks.
Many workloads today are multimedia applications which load millions of blocks of data and use
them only once. Some data are placed in cache and never reused again until they are evicted
from cache. This is especially true in the LLC, since exploitation of temporal locality in high
level cache means an inversion of temporal locality on LLC; in other words, data that is accessed
frequently will always hit on level 1 caches, thus remain unused on levels 2 and 3. Study shows
that numerous data blocks are allocated to the last-level cache and never reused or accessed
again. Those blocks that are placed in the LLC and never accessed should never be allocated on
cache to waste precious cache space. The method of not allocating some data to cache is called
cache bypassing.
Most bypass techniques have relied on ad hoc methods such as counters and tables which cannot
tackle the complexity of multicore workloads. In this work, we propose an alternative method to
predict cache bypassing using Support Vector Machine (SVM) models. Based on access traces
obtained from representative benchmarks running on the Multi2Sim simulator, supervised SVM
training was performed in order to obtain a bypass prediction model suitable for LLC in multicore processors. The SVM outputs bypassing classifiers which are integrated on the simulator to
quantify LLC performance improvements. Results show that, with appropriate parameters and
kernel functions, SVM is capable of generating bypassing models which improve LLC
performance on multicore processors, achieving an average 5.34% hit rate improvement across
SPLASH2 benchmark combinations.
keywords: machine learning, last level cache, cache bypassing
Security in Embedded Systems

Jeff Zadeh, Ph.D.
Professor, Faculty of Informatics
Burapha University
Chon Buri, Thailand
Abstract: Today more functionality is embedded into smaller devices. We need to concern
about security aspect of these devices. An embedded system is a computer system with a
dedicated function within a larger system, often with real-time computing constraints.
Embedded systems usually are designed to perform some specific task, rather than be a generalpurpose computer for multiple tasks. Embedded systems control many devices in common use
today. Most of all microprocessors are manufactured as components of embedded systems. For
example, intelligent techniques can be designed to manage power consumption of embedded
systems, portable devices such as digital watch, city traffic light, and hybrid vehicles.We discuss
some of the security concerns of embedded systems.
Evolution-based method
for clustering of high dimensional data streams
Kitsana Waiyamai *
Department of Computer Engineering, Faculty of Engineering, Kasetsart University, Bangkok, Thailand
Abstract
In recent years, clustering of data streams has become a research topic of growing interest. The streams clustering processes
data in a single pass and summarizes them in real-time, while using limited resources. Streams clustering algorithms that support
the monitoring and the change detection of clustering structures are called evolution-based stream clustering method. Many
techniques have been proposed for clustering data streams. However, very few have proposed to monitor and detect change of the
evolving clustering structures. E-Stream is an evolution-based stream clustering method that supports various types of clustering
structure evolution which are appearance, disappearance, self-evolution, merge and split. In this work, three different techniques
are proposed and evaluated in order to improve the performance of E-Stream in terms of both clustering quality and execution
time. First, to deal with high-dimensional data streams, a projected clustering mechanism is proposed to determine appropriated
subset of dimensions for each cluster. Second, the idea is to exploit background or domain expert knowledge as constraint for
improving the clustering process. Two types of instance-level constraints, Must-link and Cannot-link, are introduced and
integrated into E-Stream. Third, an online-offline-phases model is added into E-Stream. In the online phase, E-Stream is able to
support different evolving clustering structure types. In the off-line phase, the affinity-propagation clustering is adopted to deliver
the final clustering without any need for user intervention.
Keywords: stream clustering; affinity propagation clustering; off-line mode clustering; clustering structure change
detection, projected clustering, constraints-based clustering; semi-supervised learning
Title : Optimization Algorithm Integrated on Adaptive Array Antenna for

Emerging Wireless Communication System
Asst.Prof. Dr. Supakit Kawdungta
Faculty of Engineering, Rajamangala University of Technology Lanna,
Chiang Mai 50300, Thailand
Assoc.Prof. Dr.Chuwong Phongcharoenpanich
Faculty of Engineering, King Mongkuts Institute of Technology Ladkrabang,
Bangkok 10520, Thailand
ABSTRACT:
This research has proposed an optimization algorithm-integrated adaptive array antenna
(AAA) for the 2.4-2.5GHz wireless communication system. In the realization of the
adaptive antenna system, the fruit fly optimization algorithm (FOA) and modified fruit
fly optimization algorithm (MFOA) were deployed in the processing unit, respectively. In
this research work, simulations and experiments were carried out with the four-element
AAAs of two configurations, i.e. the linear and planar array configurations. The
simulation and experimental results revealed that the MFOA algorithmic scheme could
determine the direction of the maximum arrival signal in an efficient and accurate manner
and also was capable of manipulating the radiation pattern in the desired direction.
Eye gaze Tracking System
At the present, eye gaze tracking system is gaining interest as a powerful system for
helping a disable person. Eye-gaze tracking is the process of measuring the eye movement
during look at different positions on the screen and was captured by the camera. The eye
gaze image was computed and transformed to the screen positions.
Therefore, how to communicate with others people. It is more challenging to develop
the modern technology to support for diseases peoples. In addition, there are many
application require eye-gaze technology to interface between the human and the computer.
An example of the eye-gaze system application such as eye-gaze tracking control a wheel
chair, an embedded system for a Surgical Microscope. If the eye-gaze tracking system can be
detecting quickly and interpreting for the eye-gaze positions on the screen more accurate, the
efficiency of the eye-gaze system will getting higher. The user can using the eye-gaze to control
the computer more accurate. It is more challenge to be using the human eye as an input
replace the keyboard or traditional input.
The higher precision of eye-gaze tracking system is required for the eye-gaze tracking
system. The eye-gaze distance estimations is conducted to computing the distant between
the users eye and the computer screen. Previously, we proposed the eye-gaze distance
estimation using the iris region eigenvalues for user-independent eye-gaze distance estimation
with using a single camera for detection the eye-gaze distance. The trained system was used
only from one eye-gaze image information and used only single camera. The test distance
estimation is 60cm to 80 cm. The experimental results obtained by linear, logarithmic, and
power regressions were 95.98%, 95.33% and 96.37% accuracy.
A New Design of Variable Fractional-Delay FIR Filter

and Its Low-Complexity Filter Structure
Sorawat Chivapreecha
Department of Telecommunication Engineering, Faculty of Engineering,
King Mongkuts Institute of Technology Ladkrabang, Bangkok 10520, Thailand
Email: sorawat@telecom.kmitl.ac.th
This research presents a new design method and a new filter structure for the maximallyflat variable fractional-delay (VFD) FIR filter. Since the design method and filter structure are
formulated from the so-called discrete Pascal transform (DPT) and its Pascal interpolation, the
resulting VFD filter is called Pascal VFD filter. The kth-order Pascal polynomial in the Pascal
VFD filter is used for fitting (k+1) data points. The Pascal VFD filter is a real-time VFD filter
that consists of two sections, which can be realized into the front-end and the back-end sections.
The front-end section contains multiplierless digital filters, while the number of multiplications
in the back-end section is a linear function of the order k. Since the proposed Pascal VFD filter
structure can adjust the delay parameter online, it is more suitable for real-time applications.
Moreover, the proposed Pascal VFD filter structure is also divided into two types as the cascadetype and parallel-type Pascal VFD filter based on choice of front-end section. Both two types of
Pascal VFD filter use the same number of multiplications. Consequently, Pascal VFD filter can
give the low-complexity filter structure including no transient error during online adjustment of
delay parameter.
The Electric Field Distribution of a Hybrid Rectangular and Circular Waveguide

Resonator for use in Granular Material Dielectric Measurements
Abstract
This research presents a design of a hybrid rectangular and circular waveguide resonator.
The resonator is designed specifically for the purpose of granular dielectric
measurements, which resonates at 2.45 GHz. The electric field distributions of the
proposed resonator were compared against ordinary resonators, and illustrated using a
3D computer simulation of isosurfaces and their contours and 3D projections. In the
simulation the properties of samples (paddies and soils) are loaded into each design, in
order to observe the electrical field distributions of each sample. Next, the probability
density functions (PDF) was used to determine the probability of electric field
distributions within each type of resonator. Each designed resonator was created in
order to process real experiment circumstances. The experiments for each resonator
were processed using 20 paddy samples and 20 soil samples, in order to replicate real
circumstances in an efficient manner. The percentages of moisture content (%MC) from
each design were compared against %MC obtained by dry basis methodology, used to
find errors. The results showed that the proposed resonator has less error than the
rectangular and circular resonators which are 16.546%, 39.006% and 27.732%
respectively for paddy samples, and 17.088%, 32.583% and 23.638% respectively for
soil samples. This demonstrates that the proposed resonator creates better results than
the ordinary resonators, although the sample is changed.
Keywords:
Granular material / Microwave measurement / Moisture Inspection /

Non-destructive / Resonator
Smart EV: Charging, Driving and Connecting

Werachet Khan-ngern
Department of Electrical Engineering, Faculty of Engineering,
King Mongkuts Institute of Technology Ladkrabang, Thailand
AbstractSmart electric vehicles (EVs)must composes of three main functions including detection,
communication and decision. Car revolution has been started from steam power - electric power - internal
combustion engine (ICE). Nowadays, it seems to transition from ICE era to EV era. It may take some transition
periods due to the storage energy density aspect. Lithium batteries are common used in EVs which the energy
density (kWh/liter) is one ninth of gasoline. The energy density of the Lithium battery is one of the constraint of
EV driving range. Charging and driving are two EV main performance. While the connecting is a charming of
digital era for connected car. The EV charging can be categorized into wire and wireless charging in term of
connection or be categorized into normal charging and fast charging. Charging is an issue that EV can be a
drawback. Driving range of the EV depends on the energy storage of lithium batteries. Wireless power charging
(WPC) is one of option for the energy charging convenience. The pricing of the battery can be dominated on
the whole EV priceif the large battery is applied like in Tesla car. The connecting EV is an interesting issue not
only the IOT issue, but also can be the power backup to the micro grid to maintain the stability of the electric
grid power supply. Automate driving with or without man is possible which related standards and regulations
are required.
KeywordsSmart
electric
vehicles,
charging,
driving,
connected
car,
EV
ICESIT 2016
August 5-8, 2016
Industrial Session
ICESIT 2016

Xsense Information Service Co., Ltd

Company Profile

Sense Information Service Co., Ltd is the research and development company in fleet
management platforms using GPS tracking system. Our company was established in 2003 at
Software Park Bldg. Chaengwattana Rd., Pakkred, Nonthaburi by our professional team. Our
team consists of professionals in diverse fields ranging from communications engineering,
information technology professionals and highly experienced micro controller development
team. Our ultimate gold is to research and develop products to efficiently support fleet
management business. Our strategic partnerships are Software Park Thailand, Thailand
National Electronics and Computer Technology Center (NECTEC), Thai Embedded System
Association (TESA), and Software Industry Promotion Agency (SIPA).
ICESIT 2016

KKT ECU: Advance Engine management System for

after-market Automotive
Jiranart Khaomuangnoi
KKT Embedded System Engineering Co., Ltd. Thailand
Abstract: This is an Embedded System has been developed for new cars and used cars, the
purpose is to customize the functionality of the original ECU module (Piggy-back ECU) that
setup for engine lifetime and pollution control and specific for , the system is suitable for
vehicles modified fuel system , such as LPG or for E85.
This embedded system also developed for modified car for competition (Stand-Alone ECU),
its will take over control the engine and let technicians or engineers adjust by using computer
laptop connects with the system for maximize efficiency and power output.
ICESIT 2016

Sao Denki
Abstract Air Conditioner plays important roles for human comfort. Saijo Denki developed
Internet & Smart phone connected air conditioner using own developed IoT Solution. The air
conditioner can be controlled from anywhere in the world, report its status, intelligently
analyze and inform customers more than 30 error codes. Moreover, the air conditioner can
also report power consumption of air conditioner to customer's smart phone. Integrated with
GPS technology, Saijo Denki IoT air conditioner can send notification to end-user when
approaching home in approx. 3 km. So, customer can switch on AC before reaching &
getting more comfort during extremely hot or extremely cold climate.
ICESIT 2016

LehTek InnoDev: The IoT Solution for Leaning

Legacy Manufacturers to Smart Manufacturers
Introduction
Manufacturers face constant pressure to continuously improve operational performance,

reduce costs, utilize resources, and balance inventory in times. Especially with mid-low-end
Thai manufacturers are in industrial era 2.0, 2.5 that are blocked by high variability of
systems, machines, equipments, and labor skills with limited resources, as big barriers to
operational excellence as Smart Manufacturers. The IoT technology and framework will be
lean steps to address the barriers by connecting, layering, and mining data from workers,
stations, processes, lines, and plants. To model, visualize, and analyze information for
decision maker and controller to command back with collective knowledge and intelligence.
Also, the framework will be able to integrate with existing manufacturing systems such as
ERP, WMS, SCM.
ICESIT 2016

Thailand's era of vending machine is coming

ChutimanYongprapat
Venus Supply Co., Ltd. Thailand
AbstractThere are a lot of reasons that vending machine will become more and more
popular in Thailand. One of the reasons is the global recession, when people cant make
money easily; they tend to find income from selling commodities. The first successor is the
mobile topup vending machine. Its success has raised a lot of company in this market. And
when the market is becoming full, these companies ask Whats next? and many kind of
vending machine are the answers. In the same time that IoT rise, it helps developers to
develop their systems smarter and in easier way.
KeywordsVending machine, IoT, mobiletopup
ICESIT 2016

SCG Smart Living

Geasee P.*, Hothongkham S.*,Chuen-Im S.*,Sangthong S.* and Mangklabruks K.**
*Innovative Product-Technology Development, Solution Technology, SCG Building Material Co., Ltd.
**Technology Planning, Solution Technology, SCG Building Material Co., Ltd.
Abstract: According to the business influences on our Company caused by the world's
megatrends of global warming, labor shortage, smart automation and growth of elder
population worldwide, including Thailand, SCG started to set up research facilities called
SCG Housing Technology Center in 2008 at National Science Park area. Three research
areas of smart-eco-care are focused. Six years later, we selected several prominent
technologies to construct a unique prototype house named The NEST: The Next Eco
Sustainable Technology. Smart automation technology was integrated with many other
products to enhance the housing performance, safety, security and convenience functions of
The Nest. Active AIRflowTM was developed to increase the quality of air ventilation of the
house and allow the air to be ventilated throughroofing tile ventilators by using smart logic
control with cloud operating platform. Automatic night light was developed for elderly people
who need special care during night time when they wake for toilet and it enabled them to
return to their bed without any injuries and better continue their sound sleep under the warm
light. Embedded logic protocol was developed in order to achieve smoothly dimmable
lighting while individual user could enjoy any activities as long as they want. Smart lighting
control switches were also one of the products being installed in The NEST mainly for
convenienceproposes. Having been testing all functions in The NEST for more than 2 years, it
can be summarized from what we found that the majority of our customers prefer to invest
their money on safety and security rather than on their convenience.
Keywords: Smart living, Automatic night light, Roof ventilation, Elder care
ICESIT 2016

Sensor Gateway Platform

Tawan Srisang, Peranitti Tuvaphanthaphiphat, Suttinun Sangngean
Prompt Tech Care Co., Ltd
Abstract Data acquisition and monitoring is one of the first processes for planning,
prediction, detection, and for further analysis. We have designed tools for data gathering
called the Sensor Gateway Platform (SG-Platform). The platform consists of three main parts:
Hardware, Server, and Client. The hardware has been designed as a general purpose sensor
gateway divided into three parts. The main processer board runs an embedded Linux. The
application board can be customized for applying user required specific sensor, and the IP
network communication channels support Ethernet, WIFI, and 3G Cellular. For the Server, we
built the software to be responsible for communicating with our hardware and client software,
for collecting and processing data, and for alarm event detection and notification. The
software is supported on both Windows and Linux platforms. For the Client, we built an
application which can be run on multi-platforms such as Windows, Linux, Mac OS for
Personal Computer or notebook, and also iOS and Android for Mobile Devices.
With this platform, we can provide many solutions for any data acquisition prior to our
customer requirement. We can detect unusual behavior, out-of-range data values, and/or
other hardware or machine malfunction. Any mentioned event detection can be sent as an
alarm to our user via E-mail, SMS, Push notification, and window pop-up from the PC
running our client software.
Keyword Sensor, Gateway, IoT, Data Acquisition, Data Logger, Monitoring
ICESIT 2016

Silicon Craft Technology Co., Ltd.

Established in 2002, SIC offers novel, customs, and standard design microchips for RFID
applications and delivers products that carry high-value added features and superior overall
systems performance.The products quality is endorsed by years of lasting partnerships.
SICs reputation has been further solidified through its experience and expertise in the design
and development of world-class foundry semiconductor manufactures of linear and mixedsignal integrated circuits.
SIC4340 is a newly launched integrated circuit (IC) for NFC Type-2 Tag with 12-bit ADC
andsensor biasing circuit.Users can connect and communicate with a diagnostic device or a
bio-chemical sensor, to read and analyze data viaa NFC-enabled smart phone.It is designed
to acquire sensors voltage response at specific time after biasing, supporting sensor with
both resistive and capacitive. Its adjustable sensor-biasing frequency generator, adjustable
voltage limiter; adjustable current source; adjustable sampling points and warm-up clock;
makes SIC4340 versatile with chemical sensors, biosensors and other kind of potentiostatic
sensors.
Keywords: RFID, Integrated Circuit, Biosensors
Biomedical Sensor for Health Monitoring
ICESIT 2016

Mobile Application Platform for Wearable Art

Raksit Chutipakdeevong
Crafity Studio Co., Ltd. Bangkok, Thailand
Email: raksit@crafitystudio.com
Patchpon Asawachiwantorngul
Crafity Studio Co., Ltd. Bangkok, Thailand
Email: patchpon@crafitystudio.com
Abstract Crafity is a mobile platform targeted to be AECs platform for personalized

clothing and accessories, we bring artists from all around the world together to give them
capabilities to produce wearable art.
Currently, Crafity is focusing on Thailand and AEC retails and made-to-order 100% premium
organic cotton t-shirt market. To achieve made-to-order capabilities, we have adopt the
DTG(Direct to garment) printing technology which can print discrete T-shirt order with
marvelous print qualities, and reduce overstocking, production line changes and product
expiration.
Moreover, Crafity is an art gallery for art and fashion enthusiast which used t-shirt as a
medium for display. Every artworks in the gallery was create by the artists in our community,
and was cherry picked by our professional to ensure only the best is offered to our
customers.
Keywords Mobile Application Platform, Premium Cotton 100% T-shirt, Artwork, DTG
.
ICESIT 2016

BAESLab: The technology solutions for growing

elderly society
Panuwat Promsiri
BAESLab Co., Ltd. Bangkok, Thailand
panuwat@baeslab.com
Introduction According to our business, BAESLab Company Limited has been the spin-o
company from Embedded Systems Laboratory by Mr.Thanong Chotisorayuth and SE-Education
Public Co. Ltd since 2013. Our mission is to be the new generation of design house in Thailand.
By our strengths and knowledges in the Embedded systems technologies, we develop the
Smartcity solutions joins hand with Saensuk municipality and other strategic partners. For the first
phase of this project, it is designed for smart healthcare under concept of Elderly care at home.
So, The elderly peoples will have their activity monitored by their families or nurses via the smart
wristband or smart necklace including the fall detection function. In the other hand, we also
develop the Advance Service Framework to compatible with the various operating systems
especially Linux over the serveral IoT Gateway devices.
ICESIT 2016
August 5-8, 2016
Technical Program
Virtual View Synthesis Using

Moving Multi-Camera Array
Su-Min Hong and Yo-Sung Ho
Gwangju Institute of Science and Technology (GIST)
123 Cheomdan-gwagiro, Buk-gu, Gwangju, 500-712, Republic of Korea
Telephone +82-62-715-2258, Fax: +82-62-714-3164
Email:{sumin,hoyo}@gist.ac.kr
Abstract Considerable improvements of information

technologies (IT) allow consumers to enjoy various forms of
multimedia services. Recently, 3D video has been established as a
new format that provides more realistic and natural experiences
to users by free-viewpoint TV (FTV). One of the main challenges
is about rendering continuous viewpoint images using color and
depth information. Thus, image interpolation, which generates
virtual view-point images, is a key part in 3D display systems. In
this paper, we propose a virtual view synthesis method for a
moving multi-camera array using stereo images. Our method
synthesizes virtual viewpoint images using both spatial and
temporal information. The proposed method consists of two parts:
depth estimation and virtual viewpoint image generation. In order
to evaluate the view synthesis method, the peak signal-to-noise
ratio (PSNR) of depth images are estimated. Our results show that
the PSNR results from the proposed method outperforms the
conventional method.
Keywords FTV; view synthesis; depth map upsampling;
moving camera array
acquire high quality depth maps [2]. To achieve this goal, first
we propose a moving camera array. This camera systems
captures multi-view color videos at 30 frame per second and
color resolution is 1280720. To capture the moving object, our
proposed multi-camera array can move left and right sides.
Second, we propose a depth map refinement method to improve
a virtual views quality. The proposed algorithms is based on
noise aware filter for depth upsampling (NAFDU) [3]. Also, we
consider reliability of color and depth pixels to solve an edge
blurring problem. Third, we propose a virtual view synthesis
method in the moving multi-camera array. In our method, we
synthesis virtual viewpoint images using both spatial and
temporal information. Once the disparity maps are determined
from left and right image sequences, we are able to synthesis a
virtual view flexibly. After that, we search the reference image
using the color differences between the virtual view and
candidate images. Then we divide the moving and static areas
differences virtual view image and reference image. Finally, we
obtain the improved virtual view images using reference
information.
I. INTRODUCTION
Virtual view synthesis is one of the most important
techniques to make high quality videos in the free viewpoint
television and three-dimensional (3D) video. Unfortunately,
efficiently making virtual view of complex real world locations
is a challenging task. Depth image-based rendering (DIBR) is
generally used to synthesize a virtual view in free viewpoint
television (FTV) and 3D video [1].
In the previous works, most efforts were focused on static
multi-camera system. Many static multi-camera systems were
presented to capture the scene and generate 3D video contents.
Most of them are composed of fixed multiple color cameras
with parallel arrangement. In this case, if an object moves
widely, the number of camera should be increased. As a result,
camera setup and computation cost of the algorithm will be
increased. Alternatively, we can reduce the number of cameras
by increasing the camera interval. However, in this case, only
sparse information can be used in the view synthesis, so that the
quality of the generated free-viewpoint image significantly
degrades and it is difficult to expect high quality of the 3D video.
The contributions of this paper include virtual view
synthesis method in the moving multi-camera array, and
II. VIEW SYNTHESIS USING MOVING CAMERA ARRAY

So far, the focus in virtual view synthesis was concentrated
on static camera array. In the static camera array, range of
virtual viewpoint image generation is limited. In particular, the
important environments are often actively in use, containing
moving objects, such as player in speed skating. In this case, the
number of camera should be increased. Therefore, the system
complexity and cost of camera system setup increase. To reduce
the complexity and cost, the number of cameras can decreased
by increasing the camera interval. But, in this case, only sparse
information can be acquired, so that the quality of generated
virtual viewpoint image significantly degrades.
To solve this problem, our proposed method using the
moving multiple camera array. By using moving multiple
camera array, we can follow an object of interest and also obtain
a dense ray information of the scene. So moving multiple
camera array can generate the virtual viewpoint images in a
wider range in comparison with the static camera array. Fig. 1
shows range of capturing space in moving multiple camera
array.
of pixels in the filter kernel. Also, we can get the mean value in
the filter kernel using (2).
1
(2)
The standard deviation is a measure of how spreads out
pixel values are. In addition, it can represent the dispersion of a
set of pixel values from the filter kernel. So, we can decide
which region is the edge by using standard deviation. Our
proposed method use the adaptive weighting functions based on
edge information. Blending function is given by Eq. (3).
Fig. 1. Range of capturing space in moving camera array
A. Depth map estimation

When we make the virtual viewpoint image, depth or
disparity information is required. And synthesized view quality
is highly depends on the depth information. In general, there are
two types of techniques for acquiring depth information of the
scene. One is based on passive range sensors and the other is
based on active range sensors, such as stereo matching. Stereo
matching can be categorized into local, global and semi-global
methods. Local methods are processed by windows based on
correlation where the disparity is assumed to be equal for all
pixels within the correlation window. In global methods, the
task of computing disparities is cast as an energy minimization
problem. Generally, global methods are computationally
complex even for low resolution images with a small disparity
range. The execution time leaves still room for improvement.
There is a third category of stereo algorithms lying between
global stereo and local stereo. These hybrid stereo algorithms
are known as semi-global matching algorithms (SGM) [4]. To
make the high quality disparity map fast, we use the SGM
method.
B. Depth map refinement
When we make the virtual viewpoint image, depth or
disparity information is required. And synthesized view quality
is highly depends on the depth information. Unfortunately,
stereo matching remains a difficult vision problem. So our
proposed method use the multilateral filter to improve quality
of disparity maps. The proposed algorithms is similar to our
previous work [5].
First of all, we compute the standard deviation in filter
kernel. In statistics, the standard deviation () is a measure that
is used to quantify the amount of variation or dispersion of a set
of data values. A standard deviation close to 0 indicates that the
data points tend to be very close to the mean of the set, while a
high standard deviation indicates that the data points are spread
out over a wider range of values. Standard deviation in the filter
kernel is defined by (1).
(1)
1
1
,x
(3)
blending function is the distribution of the 0 to 1 [3]. Here x is

the ratio between the result of Eq. (1) and the maximum
standard deviation in each kernel. The largest standard
deviation possible will be found where the largest possible gaps
are. So, we can calculate the largest standard variation very easy.
The maximum standard deviation is calculated as in (4).
min
max
(4)
where represent the spatial neighborhood around p. M

indicates the mean in the and min, max represents the
minimum depth value and maximum depth value in the . The
standard deviation distribution is normalized in 0 to 1 by
consider maximum standard deviation. Also, it can represent
how much edge information is included in each kernel. If
standard deviation is very small in some region then the x have
almost zero. Otherwise, x is the close to one when the standard
deviation is almost same with maximum standard deviation. So,
we can determine the edge in the filter kernel by using the x
value.
After that, we consider reference pixel reliability in edge
region. In the conventional bilateral filter, if the depth map and
the color image differ significantly in appearance, the use of the
color image to upsample the depth map leads to artifacts. To
reduce these artifacts, we are considering reference pixel
reliability. First, we decide the edge region when x is greater
than threshold. In the edge region, we calculate the mean value
of the color image and depth map, via the Gaussian weighted
sum of neighbors within a filter kernel.
1
(5)
||
||
After that, we choose the reference pixels in the color image
and depth map which difference between Gaussian weighted
mean value and neighbor pixel is smaller than threshold. In our
implementation, we empirically set it equal to 10.
(6)
where p denote the center pixel in the filter kernel and q

represent the neighbor pixels in the filter kernel. n is the number
So, the edges can be preserved by considering reference pixels
reliability. Assume that there are input low quality disparity

map I, output disparity map , high resolution color image
and low quality target depth map . After blending function is
determined, depth value is defined by Eq. (7).
1
(7)
If the x is increased, is close to 1 and proposed method

refined quality of input disparity map while protecting the edge
information. Otherwise, If the x is decreased, is close to 0 and
proposed method use the depth map information.
Fig. 3. Candidates for reference
After virtual view synthesized, our proposed method find the

reference image to improve quality of synthesized view. In the
conventional method, synthesized views were made by left and
right images information. In our method, we use left, right and
reference images for virtual viewpoint image generation.
Reference images means most similar view with virtual
viewpoint images in the different times.
As shown in Figure 3, candidates (red dots) for reference
image are located in the different times. If the synthesized
image(orange dot) and synthesized image are located in the
same space, they will have small color differences. So, we can
choose the reference image using the color differences. Fig. 4
illustrates color differences between candidates and synthesized
image. In this case, we choose the smallest color differences
image (red box).
Fig. 2. Result of depth map filtering
C. Virtual view synthesis

Our method makes two left and right disparity map and
warp virtual images respectively. Then, two virtual images are
summed by weighting function. Although it has heavy
complexity due to two disparity estimation process, it generates
good quality virtual image. The static camera array has two or
more static cameras in which a viewpoint corresponds to video.
Static camera array stay still in time. All cameras are
synchronized and can acquire dynamic scene from several
viewpoints. The static camera array should be set so that the
cameras can cover a required range of 3-D space.
Based on refined disparity maps over stereo image
sequences, virtual views can be synthesized at any virtual
camera position. It is represented by the ratio of base line, which
is distance between the left and right cameras. Our proposed
method using the linear interpolation to make the virtual
viewpoint images. Virtual viewpoint image I is defined by Eq.
(8).
1
(8)
where and are the left and the right images, respectively. x
and x are pixel positions, d is disparity vector and is the ratio
of baseline. In the moving multiple camera array, we can
synthesize virtual viewpoint images using conventional linear
interpolation method.
Fig. 4. Color differences between candidates and synthesized image
In the reference image, position of the scenes are same. It

means we can use the information of reference image to
improve quality of synthesized images. Note that the moving
object has different positions between reference image and
synthesized image. Therefore, we should divide moving and
static areas in images. In the previous step, we calculate the
color differences between reference image and synthesized
image. This information represent moving object area. So
moving object mask can generated by color differences
between reference image and synthesized image. As shown in
Figure 5, mask of color differences have some noise. The noise
in the generated mask is removed by dilation and erosion, so
that a better mask is obtained for the moving and static areas.
Finally, given the mask of the moving and static areas, the
moving areas are synthesized by using left and right images,
which include blending and conventional inpainting. Virtual

view in the location of the static area is generated by reference
image.
Synthesized image using conventional method have boundary noise

around structures and background. To reduce these problems, we use
left, right and reference image for virtual viewpoint image generation.
Reference image means most similar view with virtual viewpoint
images in the different times. After applying the proposed method the
quality of static areas has been improved.
Also Table 2 shows comparison of PSNR values of the 9 virtual
viewpoint images. The term peak signal-to-noise ratio (PSNR) is an
expression for the ratio between the maximum possible value (power)
of a signal and the power of distorting noise that affects the quality of
its representation. The proposed method improved quality of the
synthesized images by 5.67 dB.
TABLE II.
PSNR COMPARISON OF SYNTHESIZED IMAGES
Fig. 5. Mask of moving object
III. EXPERIMENTAL RESULTS

In order to evaluate our system, we generate a moving camera array
sequence using stop motion animation. We captured 20 frames by
moving the object gradually, while the camera array is shifted 1 cm per
frame and distance between camera positions were set 10cm. For an
objective evaluation of the proposed multilateral filtering, proposed
method is compared with the joint bilateral upsampling (JBU), the
noise-aware filter for depth upsampling (NAFDU) and the MRF-based
depth upsampling. For an objective evaluation of the depth maps, the
bad pixel rate (BPR), whose absolute difference is greater than 1, was
used.
TABLE I.
PERFORMANCE COMPARISON OF MULTILATERAL FILTERING
IV. CONCLUSION
In an ideal parallel fixed camera array, cameras are located on a
certain line which is called the baseline. Also, all the cameras have the
same distance to the adjacent cameras. In this case, range of virtual
viewpoint image generation is limited. To solve this problem, our
proposed method using the moving multiple camera array. We apply
the semi global matching and multilateral filtering to get the high
quality disparity map and synthesis virtual viewpoint images using
both spatial and temporal information. As shown in the experimental
results, we can obtain a high quality of the virtual viewpoint images
for moving camera array. For most parts of the comparison, our
proposed method showed the better results with the results of the
conventional methods.
V. ACKNOWLEDGMENT
This research was supported by Basic Science Research Program
through the National Research Foundation of Korea(NRF) funded by
the Ministry of Science, ICT & Future Planning(2011- 0030079)
VI. REFERENCES
Table 1 shows the result of the BPR(%) comparison. According to

the results, the proposed multilateral outperforms the conventional
algorithms in terms of the BPR for all the scaling factors and all the
tested images.
Fig. 6. Result of synthesized images
Figure 6 represents the virtual viewpoint image using the

conventional method and the proposed method respectively.
[1] C. Fehn, "Depth-image-based rendering (DIBR), compression,

and transmission for a new approach on 3D-TV," in Proc. SPIE
Conf. Stereoscopic Displays and Virtual Realistic Systems XI, vol.
5291, pp. 93-104, 2004.
[2] T. Yokoi, N. Fukushima, and T. Yendo, Novel view synthesis
for dynamic scene using moving multi-camera array, in Proc.
SPIE Conf. Stereoscopic Displays and Virtual Realistic Systems
XXII, vol. 7863, pp. 24-30, 2011.
[3] D. Chan, H. Buisman, C. Theobalt, and S. Thrun, A Noise-aware
Filter for Real-time Depth Upsampling, ECCV Workshop on
Multi-camera and Multimodal Sensor Fusion Algorithms and
Applications, pp. 112, Oct. 2008.
[4] H. Hirschmueller, Accurate and Efficient Stereo Processing by
Semi-Global Matching and Mutual Information, in Proc. IEEE
CS Conf. Computer Vision and Pattern Recognition, vol. 2, pp.
807-814, 2005.
[5] S. M. Hong and Y. S. Ho, Adaptive Weighted Upsampling of
Depth Map using Edge Information, in Proc. Korean Institute
Smart Media vol. 4, no. 1, pp. 146-149, 2015
Relayed Message Sending for MQTT Protocol

Chontawee Wongpromrat
Watchara Chatwiriya
Department of Computer Engineering

Bangkok, Thailand
58601091@kmitl.ac.th
Department of Computer Engineering

Bangkok, Thailand
watchara.ch@kmitl.ac.th
Abstract Nowadays, many protocols are used in Internet of

Things (IoT) system. Message Queue Telemetry Transport
(MQTT) is one of adopted the communication between devices
and server in IoT ecosystem. To solve the "repeatedly sending the
data to the broker" problem, MQTT use the Quality of Service
(QoS). However, in some critical system, it is important to resend
the data again without or with the shortest waiting time. This
paper will propose concept and solution for solving this problem
for removing or decrease the uncertain time that have to wait in
QoS process.
Keywords IoT, MQTT, HTTP, ARP
I. INTRODUCTION
IoT is more practical in many application areas because an
increasing number of smart devices and ease of Internet
accessibility. Even though the concept of IoT is defined, but
the communication protocols to be used with are still in the
developing and competing to be standard [12].
One of the proposed protocols is MQTT, a lightweight
protocol with many advantages [1]. MQTT use publishsubscribe pattern in which the server (This server is called a
broker) uses with message intermediate [1,2]. In this pattern, a
broker will receive the data from devices, and forward to
connected clients, as listed in topic of devices. The broker will
forward data to that client only if the client subscribes common
topics.
In MQTT standard [1], QoS is defined as the service level
agreed between sender and receiver for message sending, with
the guarantees of delivering messages. Actually, the sender has
to keep sending until it gets the response, to meet QoS
conditions. With restrained bandwidth and noisy environment,
the waiting time for this pattern is unpredictable and
ineffective.
In this paper, we proposed a solution to solve this problem,
assumed there is nearby communication devices or other
communication channel. HTTP and ARP Protocol can apply by
sending data to nearby devices and retransmit data to a broker
instead of ordinary sender using.
The rest of the paper is organized as follows. Section 2
describes background knowledge. Section 3 discusses about the
concept of the proposed solution. Section 4 explains about
implementation. Section 5 describes the experiment. Lastly,
section 6 concludes the work.
II. BACKGROUND
IoT communication protocols are developed to meet
various requirements such as support heterogeneous hardware,
capable of data management, support application developers,
extend the different platforms for the formation of ecosystems
and the availability of dedicated marketplaces to the IoT [11].
One of the popular protocols is MQTT. The communication
model in MQTT can be described as sending messages
between brokers and senders, with QoS mechanism to prevent
disruption in the transmission. The QoS is described as three
levels as the following [13]:
QoS level 0 sends the message only once following the
message distribution flow, and does not check whether
the message arrived to its destination. Therefore, in case
of sizable messages. It is possible that the message will
be lost when any kind of loss comes in the way.
QoS Level 1 sends the message at least once and checks
the delivery status of the message by using the status
check message, PUBACK. However, when PUBACK
is lost, it is possible that the server will send the same
message twice, since it has no confirmation of the
message being delivered.
QoS level 2 passes the message through exactly once
utilizing the 4-way handshake. It is not possible to have
a message loss in this level, but due to the complicated
process of 4-way handshake, it is possible to have
relatively longer end-to-end delays.
Fig. 1. Packet transmission method about QoS level [13]
We note that the QoS process needs to send the message

repeated in some situation, if the connection is unstable, it will
difficult to predict the communication time and can will
consider as a serious factor in critical systems.
Moreover, the system cannot use only MQTT protocol to

solve this problem because the limitation of the protocol. So
we have to use another protocol for sending data to nearby
devices. We will use response time to measure performance as
suggested in [9].
III. CONCEPT
In the previous section, we describe a basic QoS process in
MQTT protocol. In terms of performance, there is a research
[13] which analysis about a delay and message loss problem in
the QoS process of MQTT protocol but proposed no solution.
When nearby devices receive HTTP messages, it will

forward the message to the broker and wait for the
result until time is out. But if it cannot receive
forwarded the message, primary sender will skip this
nearby device and forward the message to the others.
If the nearby device can forward the message, it will
return success result to the primary sender before
primary sender will finish the process. On the other
hand, primary sender will forward the message to other
nearby devices.
Assume that there are other communication channels beside

the problematic channel, we can propose the solution with the
following scenario:
From Fig. 2, if Sender 1 cannot send message to the broker
because of the connection problem, it usually uses a QoS
process for sending messages again and again. If the waiting
time is more than the threshold, Sender 1 will forward
messages to other sender via the alternative channel, assumed
with more reliable communication with the client, can
successfully transmit the message.
Fig. 3. Proposed state diagram of primary sender
Fig. 2. Proposed concept of solution
However, in this scenario, we need to find the nearby

devices. Then the Address Translation Protocol or ARP
protocol can be useful [3,4].
In order to forward messages to nearby devices. we apply
HTTP protocol, because it uses a lot in present day. Moreover,
the HTTP can embed in embedded devices. This approach can
be called Web of Things or WoT [5,6].
The modified scenario can be described as the following
steps (as Fig. 3 and Fig. 4):
Primary sender (we call the ordinary sender as primary
sender) will send messages to the broker. If it cannot do
this process, it will find nearby devices and keep device
addresses to list. On the other hand, primary sender will
finish the process.
Fig. 4. Proposed sequence diagram of solution mechanism
Primary sender will forward messages to each nearby

device from address list respectively.
IV. IMPLEMENTATION
In this section, we illustrate the network diagram of system
implementation as Fig. 5. If one of the network is not working,
primary sender can transmit messages to nearby devices which
will forward the messages to the broker server.
devices can forward the message, primary sender will back to

the first process again (pseudocode shows in Fig. 8).
# Third Process Forward message to each nearby
device and wait for the result
for eachNearbyDeviceIp in allNearbyDeviceIp:
result = http.send(NearbyDeviceIp[i], topic, data)
if result == Success:
stopAllProcess()
backToFirstProcess()
Fig. 8. Show pseuocode about HTTP transmitting process
When a nearby device receives an HTTP message from

primary sender, it will forward that message to a broker server.
A result from this process will return to primary sender
(pseudocode shows in Fig. 9).
Fig. 5. Network diagram of this experiment
In the first process, the primary sender tries to send message

to the broker. If it cannot send this message, it will find nearby
devices for forwarding the message (pseudocode shows in Fig.
6).
# First Process Try to send message to the broker
result = mqtt.send(server.ip, topic, data)
# Nearby Device Process Receive HTTP message

and forward it to the broker
message = http.receive();
topic = message[0]; # Receive topic of MQTT message
data = message[1]; # Receive data of MQTT message
result = mqtt.send(server.ip, topic, data);
returnSuccessResultToPrimarySender()
stopAllProcess()
else:
else:
returnFailResultToPrimarySender()
findNearbyDevice()
Fig. 6. Show pseudocode about condition of MQTT sending process
In finding nearby devices process, primary sender will use

the ARP protocol for asking IP Address of nearby devices and
store it to array list (pseudocode shows in Fig. 7).
# Second Process Get IP of nearby devices from
ARP
Fig. 9. Show pseudocode about data retransmitting process of nearby devices
V. EXPERIMENT
In this section, we show the experimental setup and the
result. By we set up the system as the following:
Open sources which we choose in this experiment are
Mosquitto as a broker server and Paho as clients.
There are 1 broker, 1 receiver, 1 primary sender, only 1
nearby device which can forward messages to the
broker and 10 other nearby devices in the network.
allNearbyDeviceIp = ARP()
Fig. 7. Showpseudocode about store IP Address to array list
The network topology is same as Fig. 5.

In forwarding message process, primary sender will
forward data to each nearby device using the HTTP protocol
and wait for receiving a result from that nearby device. If that
nearby device can forward this message, primary sender will
finish all processes. On the other hand, primary sender will
forward message to the others. But if there are no nearby
Size of payload is 1,000 bytes. (It is minimal payload in

the experiment of S. Lee, H. Kim, D.K. Hong and H.
Jus research [13])
For HTTP connection, we set the threshold time to 10
seconds. (Recommended default timeout in experiment
of [14])
For QoS 1, QoS 2 levels and the proposed solution, we

repeat the measurement for 10 times.
We measured transmission time by creating network
congestion between primary sender and broker for 10, 20 and
30 seconds. We have compared proposed solution with
transmission time obtained by each QoS level. We found that
transmission time from proposed solution is less than every
QoS level when delay is longer than 10 seconds (as shown in
Fig. 10, 11 and 12).
Fig. 10. Result when network congestion has occurred for 10 seconds
Table 1 summarizes the average transmission time and

quantity of forwarded devices. Analysis results in our
experiments show that transmission time and quantity of
forward devices has a strong positive relationship for the
proposed solution.
TABLE I.
SHOW THE AVERAGE RESULT OF EXPERIMENT
QoS 1
QoS 2
Proposed
Solution
10
10.81
11.01
14.88
Average
quantity of
forwarded
devices (unit)
7.2
20
20.91
21.01
14.83
7.0
30
30.91
30.81
14.93
7.4
Network
congestion
time (second)
Average response time (second)
In the result, the proposed solution has stable response time

than the QoS mechanism because it does not use congestion
channel for sending messages. But it is necessary to use more
process than QoS mechanism (find nearby devices by ARP,
forward message to each nearby device by HTTP and wait for
the result from each nearby device). So these processes affect
response time in the proposed solution, whereas response time
in QoS mechanism depends on network congestion time.
VI. CONCLUSION
In this paper, we propose the solution for solving delay and
message loss problem under QoS process in MQTT protocol.
Experimental results suggest that this solution can reduce
transmission time when the system has a long delay. This
transmission time is significantly associated with quantity of
forwarded devices. If we allow nearby devices to forward
message, we will be able to build a more effective system.
Regarding future studies, we will experiment under various
priorities of message for performing the experiment in a more
realistic scenario. Also, we will analyze launch time of new
added message under different payload.
REFERENCES
Fig. 11. Result when when network congestion has occurred for 20 seconds
[1]
[2]
[3]
[4]
[5]
[6]
[7]
Fig. 12. Result when when network congestion has occurred for 30 seconds
Advancing
open
standard
for
the
information
society,
https://www.oasisopen.org/
U. Hunkeler, H.L. Truong and A. Standford-Clark, "MQTT-S - A
publish/subscribe protocol for Wireless Sensor Networks," 3rd
Communication Systems Software and Middleware and Workshops,
Bangalore, IEEE, 2008, pp. 791-198.
RFC Editor, https://www.rfc-editor.org/
Y. Qin, Y. Zhou, and B. Liu, "A High Performance ARP Lookup
System for Gigabit Ethernet," 2nd Communications and Networking in
China, Shanghai, IEEE, 2007 pp. 82-86.
V. Gupta, R. Goldman, and P. Udupi, "A Network Architecture for the
Web of Things," 2nd International Workshop on Web of Things, San
Francisco, ACM, 2011, article. 3.
] C. Zhang, C. Cheng, and Y. Ji, "Architecture Design for Social Web of
Things," 1st International Workshop on Context Discovery and Data
Mining, Beijing, ACM, 2012 article. 3.
B. Aziz, "A Formal Model and Analysis of the MQ Telemetry Transport
Protocol," 9th International Conference on Availability, Reliability and
Security, Fribourg, IEEE, 2014 pp. 59-68.
[8]
A.J. Coddington, "Message Queuing for Network Monitoring,"

Technical Report, The University of Waikato, 2012
[9] E.G Davis, A. Calveras, and I. Demirkol, "Improving Packet Delivery
Performance of Publish/Subscribe Protocols in Wireless Sensor
Networks," Barcelona, Sensor, 2013, pp. 648-680.
[10] D. Thangavel, X. Ma, A. Valera, and C.K.-Y. Tan, "Performance
Evaluation of MQTT and CoAP via a Common Middleware," 9th
International Conference on Intelligent Sensors, Sensor Networks and
Information Processing, Singapore, IEEE, 2014, pp. 1-6.
[11] J. Mineraud, O. Mazhelis, X. Su, ans S. Tarkoma, "A gap analysis of
Internet-of-Things platforms," Preprint, arXiv, 2015
[12] A.P. Castellani, M. Bui, P. Casari, M. Rossi, Z. Shelby and M. Zorzi,

Architecture and protocols for the Internet of Things: A case study,
IEEE International Conference on Pervasive Computing and
Communication, Mannheim, IEEE, 2010, pp. 678-683
[13] S. Lee, H. Kim, D.K. Hong and H. Ju, Correlation Analysis of MQTT
Loss and Delay According to QoS Level, 2013 International
Conference Information Networking, Bangkok, IEEE, 2013, pp. 714717
[14] B. Liu, G. Abdulla, T. Johnson and E.A. Fox, Web Response Time and
Proxy Caching, Technical Report, Virginia Polytechnic and State
University, 200
Image Fusion using Image Blocks and Modified

Discrete Wavelet Transform
Wisarut Chantara and Yo-Sung Ho
School of Electrical Engineering and Computer Science
Gwangju Institute of Science and Technology
123 Cheomdangwagi-ro, Buk-gu, Gwangju 61005, Republic of Korea
{wisarut, hoyo}@gist.ac.kr
Abstract In digital photography, all the structures of normal

digital cameras have the finite depth of field problem.
Photographers could only focus with one point. This means the
objects apart from the focus range appear with blurring effect. In
order to solve this problem, this paper presents an image fusion
technique to extend the depth of field. Image blocks and modified
discrete wavelet transform is applied to source images. The
method is operated with DWT coefficients. The approximation
coefficients are applied with spatial frequency and selected based
on a maximum sharpness focus measurement, while the detail
coefficients are applied with sum-modified-Laplacian and chosen
based on a maximum information based fusion scheme is
proposed. The performance of the proposed method was
conducted and compared with conventional methods.
Experiment results show that the proposed method can extend
the depth of field considerably.
selection of image blocks from source images to construct the

fused image using spatial frequency, while Huang et al. [5],
evaluated focus measurement in multi-focus image fusion.
This paper proposes an image fusion algorithm by
measuring focal regions and combining in-focused parts to the
complete image. The final image displays sharply all-focused
image and shows the depth of field can be extended
exceedingly.
II.
PROPOSED METHOD
In this paper, the proposed method is used to measure infocused regions and combine these regions to all-focused
image as shown in Fig. 1. The following techniques are
involved in the proposed methods implementation.
Keywords-Depth of field; image blocks; discrete wavelet

transform; spatial frequency; sum-modified-Laplacian
I.
INTRODUCTION
Image fusion is the process producing a single image from

a set of source images, such as multi-focus, multi-sensor and
multi-frame images. The fused image has more complete
information which is more useful for human or machine
perception. Image fusion includes important application in
many different fields such as medical imaging, satellite
imaging, microscopic imaging, computer vision and robotics.
The advantages of image fusion improve reliability and
capability by redundant information and complementary
information respectively. Then, image fusion algorithm
provides an all-focused image (also referred to as an extended
depth-of-field image).
In the past, many fusion techniques and algorithms for infocused detection and fusion have been proposed to obtain allfocused images. Zang et al. [1], proposed a way of medical
image fusion technique according to wavelet transform. The
method used different characters of wavelet theory, enactment
threshold to wipe off the noise and the convolution process.
Liu et al. [2], presented multi-focus image fusion strategy
based on the lifting scheme of wavelets gives substantial
information in the fused image. Desale et al. [3], proposed a
study and analysis of image fusion technique by way of PCA,
DCT and DWT. Li et al. [4], introduced a method based on the
Figure 1. Proceduce of the proposed method
A. Block partition stack images

We capture stack source images using the Lytro camera [6].
A set of images is same viewpoint with different focus [7].
After that we partition the stack source images into blocks (88
blocks). It is clear that this method can avoid the problem of
shift-variant [5].
B. Discrete wavelet transform
Discrete wavelet transform (DWT) converts the stack
source images from the spatial domain to frequency domain.
According to Fig. 2, the image is divided by vertical and
horizontal lines. The first-order DWT output contains the
approximation coefficient (CA) or low-frequency coefficient
and the detail coefficients (CH, CV and CD) or high-frequency
coefficients. CA is sensitively with human eyes [8] while CH,
CV and CD have more detail information about the image. The

DWT can minimize color distortion.
(4)
Where TSML is a threshold value of SML and I(i, j) is the

detail coefficients at position (i, j).
C. Image fusion process
The approximation and detail coefficients of DWT
structures are combined as:
Figure 2. DWT Decomposition
1) Approximation coefficient
Approximation coefficient is the source image at the
coarser resolution level, which can be considered as a
smoothed and subsampling version of the source image. Most
information of the source image is kept in the approximation
coefficient. Then, we utilize the spatial frequency (SF), which
indicates the overall active level in an image and originated
from the human visual system, on the approximation
coefficient. The SF is expressed as:
(1)
Where RF and CF are the row frequency and column

frequency respectively:
(2)
(5)
Where Fi, j is a fused coefficient in a final image at position

(i, j). While
is a maximum coefficient information at
position (i, j).
is DWT coefficient information at t stack
source image at position (i, j). N is a set of stack source images.
D. Consistency verification
After combining the fused DWT coefficient, we verify the
result by a consistency verification process (CV). If the
considering pixel is to come from source image A, but with the
majority of its surrounding pixels from B, this pixel will be
changed to come from B. A majority filter is implemented with
3 3 window size.
E. All-focused Imaging
Finally, we perform the inverse DWT on the combined
wavelet coefficients, which are obtained by the CV process
from the previous step. Then, an all-focused image is
reconstructed into a final image.
(3)
For MN image with the approximation coefficient f(i, j) at

position (i, j).
III.
EXPERIMENT RESULTS
In this section, we present the outcomes of the proposed

method with different sample images. We conducted
experiments on sample images of size 512 512 pixels. The
experiment parameters are assigned as: image blocks = 8 8
blocks, SML window size = 3 3 pixels, TSML = 10 and CV
window size = 3 3 pixels.
2) Detail coefficients
The detail coefficients contain the detailed information of
an image, which usually have large values correspond to sharp
intensity changes and preserve salient information in the image.
Moreover, The energy of the detail coefficients of an infocused image is much larger than that of a blurred image.
Then, we apply sum-modified-Laplacian (SML) to the detail
coefficients. It can effectively represent the salient features and
sharp edge of the image. The SML is expressed as:
Figure 3. Experimental sample images
To prove our proposed method performance, the

comparative methods (such as pixel averaging, DWT and SF)
are compared with the proposed method. Fig. 4 and Fig. 5
illustrate the image fusion result using different methods.
methods as shown in Table I. The quality of fused image would

be better with higher MI and PSNR values.
TABLE I.
Images
Figure 4.
Figure 5.
PERFORMANCE OF DIFFERENT FUSION METHOD
Criteria
Pixel
Avg.
DWT
SF
Proposed
Method
MI
4.147
3.943
4.453
4.608
PSNR(dB)
26.994
24.357
28.232
28.310
MI
7.155
7.598
7.113
7.818
PSNR(dB)
24.542
21.324
24.517
24.831
IV.
Figure 4.
(Top-left) Pixel averaging, (Top-right) DWT, (Bottom-left) SF,

(Bottom-right) Proposed method
CONCLUSIONS
In this paper, we propose an image fusion method using

image blocks and modified discrete wavelet transform. To
improve the DWT algorithm, we present the adapted DWT
method using spatial frequency and sum-modified-Laplacian.
As a result of that we can correctly find out and merge the focal
regions to all-focused image. From the experiment results, we
can confirm that our proposed method properly working well
and show that the depth of field can be extended greatly.
ACKNOWLEDGMENT
This research was supported by Basic Science Research
Program through the National Research Foundation of Korea
(NRF) funded by the Ministry of Science, ICT & Future
Planning (2011-0030079).
REFERENCES
[1]
[2]
[3]
[4]
[5]
Figure 5.
(Top-left) Pixel averaging, (Top-right) DWT, (Bottom-left) SF,

(Bottom-right) Proposed method
From the criterion evaluation, the mutual information (MI)

and the peak signal-to-noise ratio (PSNR) are considered with
respect to the reference images. The proposed method provides
effective results when compared with other conventional
[6]
[7]
[8]
H. Zhang and X. Cao, A way of image fusion based on wavelet

transform, IEEE 9th International Conference on Mobile Ad-hoc and
Sensor Networks, pp. 498-501, 2013.
L. Liu, H. Bian and G. Shao. An effective wavelet-based scheme for
multi-focus image fusion, IEEE International Conference on
Mechatronics and Automation, pp. 1720-1725, 2013.
R. P. Desale and S. V. Verma, Study and analysis of PCA, DCT &
DWT based Image Fusion Techniques, Int. Conf. on Signal Processing,
Image Processing and Pattern Recognition, pp. 6669, 2013.
S. Li, J. T. Kwok, and Y.Wang, Combination of images with diverse
focuses using the spatial frequency, Inf. Fusion, Vol.2 (3), pp. 395403,
2001.
W. Huang and Z. Jing, "Evaluation of focus measure in multi-focus
image fusion," Pattern Recognition Letters, Vol.28 (4), pp. 493500,
2007.
Lytro, The lytro camera, https://www.lytro.com/.
N. Patel, lfptools, http://github.com/nrpatel/fkptools/.
C.H. Lee and Z.W. Zhou, "Comparison of image fusion based on DCTSTD and DWT-STD," Int. Multi-Conf. of Engineers and Computer
Scientists, Hong Kong, pp. 720-725, 2012.
Educating Model-based Development in

the 9th TESA Top Gun Rally
Supachai Vorapojpisut
Department of Electrical and Computer Engineering
Thammasat University
Pathumthani, Thailand
vsupacha@engr.tu.ac.th
AbstractThis paper presents the education of model-based

development via a project-based learning approach at the 9th
TESA Top Gun Rally. By dissecting the scope of learning, the
learning objective was mapped based on simulation techniques
and development activities. Then electronic throttle control was
chosen as the problem with STM32F4 boards as hardware
platform. Finally, lesson learned from the hosting of event is
discussed.
Keywordsmodel-based development; project-based learning;
MATLAB/Simulink; electronic throttle control
I. MODEL-BASED DEVELOPMENT
Model-based development (or MBD in short) [1] is an
approach for developing embedded software using system
models as development artifacts that link from requirement
phase to testing phase. Representing architecture and
behaviors as models, developers can apply simulation tools to
validate and verify system operations early in the development
process. Consequently, several industrial sectors acknowledge
the necessity and advantages of MBD approaches in their
software projects especially those with safety critical
requirements. For example, automotive software development
has migrated to the MBD approach in order to cope with
expanding scope and complexity of software functions.
The MBD approach differs from the tradition approach in
several aspects. The flow of conventional software
development starts by analyzing software architecture and
functions from text-based documents, writing code to fulfill
features, and testing code according to specifications in the
original documents. Such flow demands engineers and
developers with different technical backgrounds and skill sets
to work together. While the MBD approach starts by modeling
functions and behaviors from specifications stated in the textbased documents, then models will be used and refined
throughout development activities. Using simulation and code
generation tools, simulated results and execution image can be
obtained for verification and hardware deployment,
respectively. Hence, timing behavior and computation
accuracy can be evaluated before launching a field trial.
At present, the main obstacle in applying the MBD
approach is differences in development tools, development
activities, and fundamental concepts. Several knowledge and

skill sets are required to fulfill such technical gaps. In the case
of automotive software development, modeling and simulation
of mechatronic systems can be identified as the necessary
knowledge. While the usage of MATLAB/Simulink is one of
required skill sets due to its majority in the field of automotive
software development. It should be noted that such knowledge
and skill sets are still missing from almost engineering
curriculums in Thailand.
The education of MBD can be separated into three parts,
namely development process, methodology, and development
tools. Reviewing existing MBD courses, the ratio of education
contents depends on learning objectives, hardware platforms,
and course periods. A project-based learning at University of
Adelaide applied the MBD approach to design, build, and
control an electric diwheel [2]. The project has spanned three
years to model plant and controller, visualize animated
motions with simulation, and generate code for the embedded
controller. University of Michigan offered the EECS461
Embedded Control Systems course [3] for students to learn
and practice the modeling and implementation of the haptic
wheel experiment.
Thammasat University is one of six education institutes
that joined the Thailand Electronics Design Industry Human
Resource Development Project supported by the National
Science and Technology Development Agency (NSTDA) and
Thailand Board of Investment (BOI). One target of the project
is to promote and support the education of skilled engineers
for automotive industry. As an activity of the project,
Thammasat University has collaborated with Thailand
Embedded Systems Association (TESA) to host the 9th TESA
Top Gun Rally (TGR) event. The learning objective has been
set to the education of fundamental MBD concepts and skills.
In the following, concepts, decisions, and experience gained
from hosting the event are explained.
II. COURSE DEVELOPMENT
TESA Top Gun Rally is one of major academic events in
Thailand which aims to educate and promote embedded
system technology. The event is organized as a skill
development camp in which participants will learn and
practice via a project-based learning approach.
This work is supported by Thailand Embedded Systems Association and

the research funding of Faculty of Engineering, Thammasat University.
A. Learning Objective
Embedded software development usually relies on the Vmodel development process [4] which abstracts development
activities into the specification stream (requirement, design)
and the verification stream (testing) with coding as the joint
activity between both streams. The V-model can be extended
for the MBD approach by considering modeling and
simulation techniques for each phase of development.
1.
Model-in-the-Loop Simulation (MILS): all models are

simulated within simulation environment.
2.
Software-in-the-Loop Simulation (SILS): designated

models are converted into source code and built into
an execution module in order to validate operations
outside simulation environment.
3.
Processor-in-the-Loop Simulation (PILS): designated

models are converted into source code and deployed
into target hardware in order to validate operations
inside hardware.
4.
Rapid Control Prototype (RCP): models are interacted

with physical environment via I/O interfaces.
5.
Hardware-in-the-Loop Simulation (HILS): designated

models are deployed into target hardware. Then the
hardware will interact with simulation environment
via I/O interfaces on computer.
Fig. 1 illustrates the extended V-model that relies on MILS for

the proof-of-concept phase, SILS and PILS for the prototyping
phase, and HILS for the pre-production phase.
TABLE I.
MILS
LEARNING SCOPE
SILS
2.
Controller analysis and synthesis.
3.
Offline and real-time simulation.
4.
Hardware deployment.
Due to the time limitation, the MBD methodology, i.e.

modeling and simulation of plant and controller was selected
as the emphasized learning goal. While the familiarity of
MBD tools and the understanding of MBD process are to be
achieved by spreading assignments across development
activities as shown in Table I.
HILS
Simulation
Deployment
B. Problem Setting
Corresponding to the automotive industry, the problem is
selected to be the electronic throttle control (ETC). The ETC
problem offers several advantages with respect to its
complexity layers. The simplest layer is to design a feedback
controller for the angular position control of DC motor. The
design problem becomes more complicated by considering
operating modes, e.g. failure modes and cruise control.
Fig. 2. Throttle body.
The basic control structure of ETC consists of four parts

including throttle motor as plant, pedal position sensor (PPS)
as input, throttle position sensor (TPS) as output, and engine
control unit (ECU) as controller. The operation of ECU is to
perform feedback control by comparing between TPS and PPS
voltages, then generate PWM signal to drive throttle motor.
The actual ETC structure as shown in Fig. 3 will also consider
additional information from air flow sensor and other ECUs in
order to achieve better engine performance.
For MBD of embedded control systems, each development

phase can be manifested into four steps:
Plant modeling.
RCP
Controller design
Fig. 1. An extended V-model for MBD.
1.
PILS
Plant modeling
Fig. 3. Electronic throttle control system.
C. Hardware Platform
The hardware platform for the MBD education consists of
two processor boards and an I/O board as shown in Fig. 4. The
processor board is the STM32F4 Discovery board [5]
equipped with STM32F407VG, 32-bit ARM Cortex-M4
microcontroller running at 168MHz. This board was chosen
for its availability of code generation tool, that is the Waijung
library [6].
TABLE II.
Day
1st
2nd
3rd
4th
5th
Fig. 4. STM32F4 Discovery board and I/O board.
The I/O board has been designed for required features of

HIL Simulator, including pedal (PPS) and throttle body (motor
and TPS) interfaces. In addition, CAN transceiver chip
(MCP2515) is added to support the imitation of vehicle
network. One STM32F4 Discovery board will be attached as
the processor part of the Simulator. Fig. 5 illustrates the
structure of the I/O board.
SCHEDULE AND CONTENT MAPPING
Activity
Topics
II
Modeling with Simulink
Control system design
Workshop
Code generation with Waijung library
Assignment
Modeling I/O interfaces
Assignment
TPS, PPS, PWM signal generation models
Assignment
Controller design and verification
Assignment
Hardware simulator
Assignment
Project demonstration and presentation
Lecture
Electronic throttle control
Workshop
Lecture
Column I of Table II denotes the target simulation

techniques to be practiced, i.e. M for MILS, S for SILS, P for
PILS, R for RCP and H for HILS. While column II gives the
target design steps, i.e. P for plant modeling, C for controller
systhesis, S for simulation, and D for hardware deployment.
Each assignment was given an incomplete model as the
starting point. Then each student team will modify the model
according to given specification. For example, a template
model for throttle dynamics is given as shown in Fig. 6 in
which the subsystem Throttle_sensor is the topic for the
modeling of sensor characteristics.
Fig. 6. Modeling assignment of throttle dynamics.
Fig. 5. Structure of I/O board.
III. TOP GUN RALLY

th
The 9 TESA Top Gun Rally was held at Thammasat

University, Rangsit Campus during January 4-9, 2015. There
are 28 student teams from 15 universities that enrolled to the
camp. Each team (3 students) was provided with one set of
hardware platform and the trial version of MATLAB 2014b.
The camp period was separating into two days for the
study of development tools and techniques and three days for
the project implementation. This results in the time limitation
of less than 30 hours for the project implementation.
Therefore, the problem was decomposed into smaller
assignments to be accomplished on a daily basis. Daily
activities are summarized in the following table.
Each team has been asked to use the information from a

technical datasheet related a throttle sensor as shown in Fig. 7.
Two voltages are generated redundantly according to the
opening angle of throttle valve for the safety purpose. Based
on their knowledge in Simulink and Stateflow, each team has
submitted their solution to accomplish the requirement. Then
the technical committee will evaluate points from the validity
and completeness of their solutions. For example, one team
has proposed to model correlated VTA1 and VTA2 signals
using NORMAL and VTA2_SAT modes as a state machine
modeled in Stateflow as shown in Fig. 8.
Fig. 7. Characteristics of throtte sensor.
further activities. There are several key observations from

daily activities.
1.
Almost teams did not have backgrounds on modeling

and simulation with Simulink.
2.
The six teams earning more than 65 points are from

electrical engineering (2), computer engineering (2),
biomedical engineering (1), mechatronic engineering
(1). Such team backgrounds are different from
previous events in which computer engineering field
dominates top spots.
3.
The earned points were higher for hardware-related

assignments (63 for code generation and 47 for signal
generation) compared to modeling and simulation
assignments (36 for Simulink and 29 for simulator).
4.
The lowest earned point in controller/simulator

assignments is due to the inclusion of failsafe and
cruise control features. The models to accomplish
these features are complicated.
5.
There was no actual demonstration of electronic

throttle control. Such absence may restrict the
understanding of MBD approach.
Fig. 8. An answer for the throttle sensor modeling assignment.
In total, the evaluation depended on 100-point distribution

over daily activities as 10, 25, 25, 30, and 10 points. Table III
shows the summary of earned points. While Fig. 9 illustrates
the point distribution of 28 participated teams. The winner
team from King Mongkut's University of Technology North
Bangkok earned 87.7 points.
TABLE III.
Day
POINT SUMMARY
Avg
Max
1st
Modeling with Simulink
3.6
36
8.1
81
2nd
Control design & code generation
15.6
63
19.9
79
Signal generation
11.6
47
22.7
91
4th
Controller and simulator
8.7
29
29
97
5th
Presentation
7.2
71
8.9
89
rd
Topics
Total
46.8
According to these observations, an MBD course is developed

at Thammasat University to extend the learning period to one
semester. The course content covers modeling and simulation
fundamentals with the coverage of Simulink and Stateflow
practices. The course has adopted the haptic wheel experiment
[3] as the problem due to its simple hardware requirement (DC
motor with position sensor).
In addition, an academic event is planning to promote the
education of automotive embedded system. The event will be
organized into two activities: one training workshop and one
project-based learning that spans one semester.
88.6
Acknowledgment
The author would like to express our deep gratitude to
event sponsors and Top Gun Rally technical committees for
their financial support and knowledge contribution.
References
[1]
[2]
[3]
[4]
Fig. 9. Overall point distribution.
[5]
IV. DISCUSSION
The 9th TESA Top Gun Rally was the first academic event
to target the MBD education in Thailand. Due to the time
constraint and technical gaps, the learning outcome as
expressed by 46.8 average points should be improved by
[6]
A. Sangiovanni-Vincentelli and M. Di Natale, Embedded System

Design for Automotive Applications, Computer, Vol. 40, Issue 10,
pp.42-51, October 2007.
B. Cazzolato, University of Adelaide Undergraduates Design, Build,
and Control an Electric Diwheel Using Model-Based Design,
Mathworks Newsletter, 2012.
J. Freudenberg and J. Cook, Embedded Control Systems Education at
the University of Michigan, Mathworks News & Notes, 2009.
Software Enginerring Center, Embedded System development Process
Reference guide. Information-technology Promotion Agency, 2012.
ST Microelectronics, UM1472 User manual Discovery kit for
STM32F407/417 lines. 2014.
A.H.Z. Farnaz, H.S. Sajith, P.J. Binduhewa, M.P.B. Ekanayake, and
B.G.L.T. Samaranayake, Low cost torque estimator for DC servo
motors, Proc. of 2015 IEEE 10th International Conference on Industrial
and Information Systems, December 2015.
Census Transform based Stereo Matching

with Gradient Stereo Images
Jaeryun Ko and Yo-Sung Ho
Gwangju Institute of Science and Technology (GIST)
123 Cheomdan-gwagiro, Buk-gu, Gwangju, 500-712, Republic of Korea
Telephone +82-62-715-2258, Fax: +82-62-714-3164
Email:{jrko,hoyo}@gist.ac.kr
AbstractIn this paper, we propose global stereo matching

method with census transform as the matching cost, not only on
the stereo intensity images. In order to get more accurate disparity
values on the area of depth discontinuity, we applied census
transform on the gradient stereo images. Determining the
matching cost, we applied predetermined weight value to both
operations of census transform. Finally we can get better results of
disparity map than the same method without gradient images
from our experiments.
Keywordsstereo matching; disparity map; census transform;
belief propagation; gradient image;
the matching cost on stereo matching. In our experiments we can

get the better results than without using gradient images.
II. GLOBAL METHOD ON STEREO MATCHING
A. Belief Propagation (BP)
Belief propagation on stereo matching is going along
Markov network which is like Fig. 1. Each node on the Markov
network, that is equal to each pixel, send the value called as
message to neighboring pixels [1]. Disparity values are assigned
to corresponding pixels with calculating messages from
neighboring pixels.
I. INTRODUCTION
Disparity map generation from stereo images is one of the
fundamental process for three-dimensional reconstruction and
has been researched into various methods. One of these methods
is called as stereo matching, which is estimating the depth of
projected view from stereo images of different views. Generally,
it is divided into two classes; one is local method and another is
global method.
Local methods on stereo matching are estimating the
disparity with comparison of the matching cost from views on
the right and the left. It is simple to implement and fast to be
suitable for real-time applications. For calculating matching
costs, the size of the matching window must be determined at
first, which has an effect on the quality of disparity map. The
disparity map from using local methods usually shows low
accuracy than global-based methods.
Global methods on stereo matching are based on Markov
random fields and estimating proper disparity values by
minimizing predefined energy functions. Energy function can be
minimized by using optimization algorithms such as belief
propagation, graph cut and dynamic programming technique.
Energy optimization process takes the relationship between the
entire pixels in the image into account, so it is slower than local
methods.
In this paper, we applied gradient stereo images with census
transform to improve the accuracy of disparity values on the area
of depth discontinuities. Gradient images contain the edge
information of original stereo images and it is related to the
depth discontinuity regions. Basically we exploited belief
propagation with intensity-based census transform to calculate
Fig. 1. Markov network in the image
In Fig. 1, disparity candidate , = 1, , has the

matching cost function. mi represents message from node yi to
node xi . mij is message from node xi to node xj . These nodes
must be initialized into uniform distribution as the first step of
belief propagation.
For the next step, the updating process iterates with two costs
for matching until the messages from neighboring nodes
converge through Eq. (1). ( ) is the matching cost of
assigning to node . We used census transform as the
matching cost ( ). ( , ) represents the smoothness cost,
which does not change smoothly on large discontinuities such as
object boundaries. For the smoothness term, we used the
truncated linear model like Eq. (2), where is the constant that
controls the rate of increase and is the upper threshold. After
the updating process, each belief bi ( ) of the corresponding
pixel is calculated by using Eq. (3). Finally the disparity value
with corresponding pixel is determined to xi which minimizes
the belief of corresponding pixel.
1
mij = min[( , ) + ( ) + () ,
] (1)
( , ) = min(| |, )
(2)
bi ( ) = ( ) + () ,
(3)
of the result images, proposed method shows better depth

discontinuity smoothness with the object at the right bottom. We
can get more accurate disparity values on the depth discontinuity
area like object boundaries as well.
B. Census Transform with Gradient Images

Census Transform uses relative intensity of input images
which performs robust under different absolute intensities of
input images and noises [2]. Function C(P) is census transform
of a pixel P in Eq. (4), where denotes concatenation, D means
non-parametric window around P and is defined by Eq. (5).
This function maps the local neighborhood of surrounding a
pixel P to a bit string representing the set of neighboring pixels,
whose intensity is less than that of P.
C(P) =[i,j] (, + [, ])
(4)
1, > + [, ]
(, + [, ]) = {
0,
(5)
Census transform converts relative intensity difference to 0

or 1 into one-dimensional vector. Two pixels of census
transformed images are compared for similarity using Hamming
distance, i.e. the number of bits that differ in the two bit strings
as shown in Fig. 2. To compute correspondence, the Hamming
distance is minimized after applying the census transform.
(a) Census transform
(b) Proposed method (c) Ground truth
Fig. 3. Census transform from windows of stereo image
IV. CONCLUSION
Fig. 2. Census transform from windows of stereo image
For improving the possibility of assigning more accurate

disparity into the pixels near the depth discontinuities, we
applied census transform to gradient images generated from
original stereo images and exploit it to calculate the matching
cost [3]. First we obtain gradient images Gx from input stereo
images by applying horizontal Sobel derivative operator. Then
same as census transform to intensity images, we get the
Hamming distance of Gx to define the matching cost with
weights like Eq. (6).
Cost(xi , ) = + (1 ) (6)
III. EXPERIMENT RESULTS
In order to compare proposed improvement method with
using census transform based stereo matching without gradient
images, we use for stereo images, Cones, Teddy, Tsukuba, and
Venus which are used in [4]. Window size of census transform
is fixed to 5 5 and the weight in the matching cost is
determined to 0.6. We use hierarchical belief propagation as one
type of belief propagation implements. Fig. 3 shows how well
the proposed method with gradient images makes just a little
better result than the ordinary one without them. In the last row
Census transform as a matching cost function usually shows

good performances for global method stereo matching with
belief propagation. In order to get more accurate disparity values
on the area of depth discontinuity, we proposed global stereo
matching with census transform on gradient stereo images. We
modified the matching cost with added Hamming distance of
census-transformed gradient images. Eventually we can get
more accurate disparity values near the depth discontinuities
ACKNOWLEDGMENT
This research was supported by Basic Science Research
Program through the National Research Foundation of
Korea(NRF) funded by the Ministry of Science, ICT & Future
Planning(2011-0030079).
REFERENCES
[1]
[2]
[3]
[4]
S. Yedidia, W. T. Freeman, and Y. Weiss, Understanding belief

propagation and its generalizations, Exploring Artificial Intelligence in
the New Millenium, pp. 239-269, 2003.
S. Patil, J. S. Nadar, J. Gada, S. Motghare and S. S. Nair, Comparison of
Various Stereo Vision Cost Aggregation Methods, International Journal
of Engineering and Innovative Technology, vol. 2, pp. 222-226, 2013.
C. Rhemann, A. Hosni, M. Bleyer, C. Rother, and M. Gelautz, Fast CostVolume Filtering for Visual Correspondence and Beyond, Computer
Vision and Pattern Recognition, pp. 3017-3024, 2011.
D. Scharstein, and R. Szeliski A taxonomy and evaluation of dense twoframe stereo correspondence algorithms, International Journal of
Computer Vision, vol 47, no. 1-3, pp. 7-42, 2002
A Real-time Extension of Simulink

Support Packages for Arduino
Tanapong Piansompon and Supachai Vorapojpisut
Department of Electrical and Computer Engineering
Faculty of Engineering, Thammasat University
Khlong Luang, Pathum Thani 12121, Thailand
ptanapong@gmail.com
AbstractThis paper presents the development of a library

that extends the Simulink Support Packages for Arduino
hardware. The library has been used to develop a prototype
for measuring EMG signals. Design objectives are to sample
signals at frequencies 256 and 1024 Hz, process signal,
transmit data via a Bluetooth at a speed of 230,400 bits per
second. To confirm real-time property, we have measured the
processing time of EMG and IMU measurements about 632
microseconds. Hence, there is available time left to process
sampled signals about 344 microseconds at 1024 Hz sample
rate.
KeywordsArduino; Real-time; Digital Signal Processing;
Model-Based Design
I.
develop computer software with signal processing capability

before measured data can be viewed. To overcome
computation requirements, this work has selected Arduino
Due platform due to its ARM 32-bit chip processor running
at 84 MHz which is much faster than AVR microprocessor.
Additionally, the Simulink-based software development
enables signal processing features from DSP System
Toolbox which are accurate and reliable compared with
conventional C coding. Then, software development is
focused on creating model to access Arduino Due hardware
and sensor modules, process measured data, and use
Simulink to generate C code and build into Arduino
firmware.
INTRODUCTION
The Arduino microcontroller board is an open source

hardware which is widely popular. Software development for
the Arduino board is developed in C or C++ with the
Arduino IDE. Arduino IDE has libraries for connecting with
hardware, and can add external libraries from developers.
But these libraries are developed for general usage with
limitations for specific applications such as measuring and
processing biomedical signals. So researchers who want to
apply Arduino for measuring biomedical signals, still need to
develop most part of software by themselves. Especially,
programming of digital signal processing is more
complicated than other programming.
Simulink is a software tool that used for model-based
development. Simulink enables the rapid prototyping of
signal processing device with Simulink Support Packages for
Arduino Hardware which is a library that help to generate
code for Arduino hardware, and DSP System Toolbox which
is a library for digital signal processing. Our research group
is interested in muscle activity measurement which involves
sampling signals from surface electrode and IMU unit.
However the library Simulink Support Packages is not
proved for its real-time property and its support of multi-rate
sampling. Therefore this work attempts to create a library
that extends for such purposes.
The Arduino platform has been used for biomedical
applications [1 - 3] based on Arduino UNO which uses AVR
8-bit microcontroller. These applications measure biosignal
then transmit the data to be processed on the computer. The
absence of on-board signal processing enforce researcher to
Fig. 1. Example model of Simulink Support Package.
This paper presents the development of an extended

library for Arduino Due for prototyping EMG and
accelerometer measuring devices. The target prototype is
those devices used for muscle activities in which EMG and
IMU signals (accelerometer, gyroscope and magnetometer)
are sampled, processed and reported to computer in realtime. An example application for such device is to collect
movement data from patients while receiving rehabilitation
[4].
This work considers real-time requirements regarding to
sampling rate of combined EMG and IMU signals. The study
of sampling rate for EMG signals [5] revealed that the
frequency range of EMG signals is 500 Hz, so the sampling
rate of EMG signal is 1000 Hz. Therefore this work uses the
fixed sampling rate of EMG signal at 1024 Hz. While the
sampling rate of IMU signals varied from 16 to 256 Hz [6 9]. This work uses the fixed sampling rate of the IMU
signals at 256 Hz. The survey of signal processing
techniques for EMG signals makes a list of FFT [10], Notch
Filter, Band-pass Filter, Adaptive Filter [11], Wavelet, Mean,
RMS, SD [12] and Butterworth Filter [13] which are
included in the DSP System Toolbox.
II.
CODE GENERATION FOR ARDUINO
The C language is suitable for coding hardware accessing

operations, but requires high skills for coding signal
processing operations. As a consequence, C-based software
development is not appropriate for rapid prototyping in
research projects involving signal processing hardware.
Model-based development eases such software development
by creating a model on Simulink consisting of signal
processing blocks, converts into source code, build image
file and deploy to hardware. To use model-based
development with an embedded processor, either Embedded
Coder which requires MATLAB Coder/ Simulink Coder or
Simulink Support Package must be installed into
MATLAB/Simulink environment.
Simulink Coder is a tool for generating C code from
Simulink diagram, state flow charts and MATLAB
functions. Simulink Coder allows source code from other
languages such as C or C++ language to be used in
Simulink model using S-Function Builder block. The Target
Language Compiler (TLC) provides build configuration for
generated code to deploy into target hardware or to simulate
in computer without hardware.
Simulink Support Package for Arduino hardware is an
extension for code generation on Arduino hardware. When
the extension is installed, Simulink will add the library and
the TLC of Arduino hardware. Then Simulink can generate
code from model to deploy into Arduino hardware. This
work uses Simulink Support Packages for Arduino Due for
code generation and basic I/O features.
Fig. 3. Configuration window of S-Function Builder block.
The development of additional blocks can be done by

providing code and configuration to S-Function Builder
block as follows.
- Block name.
- Sample time.
- Name and data type of Input and Output.
- Filename of source code.
- C code which runs only once at beginning.
- C code which runs repetitively on sample time.
S-Function Builder block will compile the above information
into MATLAB Executable (MEX) file that can be included
into generate code for Arduino hardware.
III.
LIBRARY DESIGN
This work started by studying limitations of Simulink

Support Packages for Arduino Due for biosignal
measurement. Our study showed the following
disadvantages.
1. The Analog Read block converts analog signal into 10bit digital value which is compatible with most Arduino
boards that use AVR microcontroller. But the Arduino Due
board uses an ARM microcontroller consisting of ADC with
12-bit resolution providing more accurate results.
Fig. 2. Example block of Arduino Hardware Support Package.
The library has three groups of I/O blocks. The first

group consists of blocks for accessing on-chip peripherals
such as Digital Input/Output, Analog Input/Output, Serial
Receive/Transmit and PWM. The second group consists of
blocks for interfacing with the Ethernet Shield such as
TCP/IP Send/Receive, UDP Send/Receive, ThingSpeak
Write. The third group consists of blocks for interfacing with
the Wifi Shield. All blocks in these libraries are generalpurpose.
2. Bluetooth module has CC2541 chip with UART

interface. Bluetooth module was set to slave mode while
computer was set to master mode. When Bluetooth link is
established, Serial Transmit block can be used to transmit
data via UART to communicate with computer. The
maximum baudrate is 115,200 bits per second. However if
the sample rate is set at 1024 Hz, the time to transmit data
must be reduced to increase available time to process
sampled signals. This work considers the improvement of
serial communication to the maximum speed of Bluetooth
module which is 230,400 bits per second.
3. I2C Read and I2C Write blocks transmit or receive
data via I2C bus. To use IMU modules, customized block are
needed for reading sensor values at 400,000 bits per second.
The scope of the library is designed to support biosignal
measurement from an EMG electrode and an IMU module
while signal processing features will be relied on existing

libraries of Simulink. The overall design of the library is
summarized in Table I.
TABLE I.
Block
Fig. 4. Data frame for transmitting EMG and IMU data.
LIBRARY STRUCTURE
Function
Toggle status of digital output pin
for processing time measurement
Convert analog signal to digital
value at 1024-Hz sampling rate and
12-bit resolution.
Read
data
from
3-axis
accelerometer, 3-axis gyroscope and
3-axis magnetometer in an IMU
module at 256-Hz sampling rate.
Format data frame and transmit

data via Bluetooth module with
baudrate 230,400 bits per second.
The real-time property of the library has been evaluated

by the measurement of processing time. Two issues were
considered including the time used for data transmission and
the time to read data from IMU module. The time used for
data transmission via Bluetooth module can be calculated
from the equation, as in (1).
Time (second) =
Transmission Data (bits)

Transmission Speed (bits/second)
The time to read from the IMU module (three sensors

with three-axis) each is approximately 868 microseconds.
Therefore data sampling from EMG electrode and the IMU
module is arranged into three reading modes with the
sampling rate of 976 microseconds as follows.
- Mode 0: Read EMG data only.
- Mode 1: Read EMG data and accelerometer data.
- Mode 2: Read EMG data and gyroscope data.
- Mode 3: Read EMG data and magnetometer data.
The time to read from accelerometer and gyroscope are
approximately 222 microseconds, while the time to read
from magnetometer is approximately 242 microseconds.
These confirm our system as a real-time system with respect
to 256-Hz sampling rate.
IV.
EVALUATION
The evaluation of the library began by measuring the

processing time of each block separately, measured all
blocks connected together, and then measured the processing
time for signal processing blocks. Finally, the overall
processing time of the model was measured to prove that the
system is real-time with respect to deadlines. The testing
procedure is as follows.
1. Measure test signals with 1024-Hz sampling rate and
12-bit resolution.
2. Read IMU data with 256-Hz sampling rate and 16-bit
resolution.
3. Transmit 10-bytes data packet via Bluetooth module.
4. Measure the processing time of the model with and
without signal processing blocks.
The measurement of processing time is considered from
sampling block until transmit block by specifying the priority
of Toggle_Pin blocks as the highest priority (start time) and
the lowest priority (finish time). The block priority is
configured by setting execution order of the blocks as shown
in Figure 5. Then the processing time is measured by an
oscilloscope with CH1 for digital pin of start time and CH2
for digital pin of finish time.
(1)
Combining EMG (2 bytes) and IMU (6 bytes 3) data,

the time to transmit data packet via Bluetooth module at
230,400 bits per second is approximately 868 microseconds.
Even the deadline for real-time sampling is approximately
976 microseconds, the remaining time is too short for signal
processing. Therefore, this work has compacted the data
protocol to 10 bytes by interleaving IMU data at lower
sampling rate as shown in Figure 4. The interleaves IMU
data is indicated (accelerometer, gyroscope and
magnetometer) with the stop byte. The estimated time to
transmit the data packet will be reduced to 434
microseconds.
Fig. 5. Simulink model which shows Execute Order.
[2]
[3]
[4]
Fig. 6. Processing Time of Model in Figure 5.
In Figure 6, the time to measure and transmit only EMG

signal (Mode 0) is approximately 390 microseconds. The
time to measure and transmit EMG signal and accelerometer
or gyroscope (Mode 1, Mode 2) is approximately 612
microseconds. The time to measure and transmit EMG signal
and magnetometer (Mode 3) is approximately 632
microseconds. The model is confirmed to execute every 976
microseconds, that is a real-time system at 1024-Hz
sampling rate.
[5]
[6]
[7]
[8]
[9]
[10]
[11]
Fig. 7. Simulink model for measurement processing time (include DSP).
[12]
[13]
N. M. Zainee, K. Chellappan, Emergency Clinic Multi-Sensor

Continuous Monitoring Prototype Using e-Health Platform, IEEE
Conference on Biomedical Engineering and Sciences (IECBES),
December 2014
R. B. Ambar, H. B. M. Poad, A. M. B. M. Ali, M. S. B. Ahmad, M.
M. B. A. Jamil, Multi-sensor Arm Rehabilitation Monitoring
Device, International Conference on Biomedical Engineering
(ICoBE), February 2012
N. Junchaya, B. Rungroungdouyboonb, S. Vorapojpisut,
Biofeedback for gait rehabilitation based on gait pattern variability,
The 4th International Symposium on Engineering, Energy and
Environments, November 2015
J. C. Ives, J. K. Wigglesworth, Sampling rate effects on surface
EMG timing and amplitude measures, Clinical Biomechanics,
Volume 18, Issue 6, July 2003
D. Gafurov, K. Helkala, T. Sondrol, Biometric Gait Authentication
Using Accelerometer Sensor, Journal of Computers, Vol. 1, No. 7,
October/November 2006
D. M. Karantonis, M. R. Narayanan, M. Mathie, N. H. Lovell, B. G.
Celler, Implementation of a Real-Time Human Movement Classifier
Using a Triaxial Accelerometer for Ambulatory Monitoring, IEEE
Transactions on Information Technology in Biomedicine, Vol. 10,
No. 1, January 2006
D. Gafurov, E. Snekkenes, P. Bours, Gait Authentication and
Identification Using Wearable Accelerometer Sensor, IEEE
Workshop on Automatic Identification Advanced Technologies, June
2007
J. Mantyjarvi, M. Lindholm, E. Vildjiounaite, S. M. Makela, H.
Ailisto, Identifying Users of Portable Devices from Gait Pattern with
Accelerometers, IEEE International Conference on Acoustics,
Speech, and Signal Processing, March 2005
J. Gra, P. Szecwka, A. Woczowski, "FFT Based EMG Signals
Analysis on FPGAs for Dexterous Hand Prosthesis Control", ManMachine Interactions. vol. 59, K. Cyran, et al.: Springer Berlin /
Heidelberg, 2009, pp. 655-662.
L. Zhao, P. Yuan, L. Xiao, Q. Meng, "Application of digital signal
processor in EMG-based human machine interface", Control
Conference (CCC), 2010 29th Chinese, 2010
K. Mahaphonchaikul, D. Sueaseenak, C. Pintavirooj, M. Sangworasil,
S. Tungjitkusolmun, "EMG signal feature extraction based on wavelet
transform",
Electrical
Engineering/Electronics
Computer
Telecommunications and Information Technology (ECTI-CON),
2010 International Conference on, 2010
C. J. De Luca, L. D. Gilmore, M. Kuznetsov, S. H. Roy, Filtering the
surface EMG signal: Movement artifact and baseline noise
contamination, Journal of Biomechanics 43, 2010
Fig. 8. Processing time of Model in Figure 7.
The evaluation of signal processing blocks was

performed as shown in Figure 7. The available time for
processing accelerometer or gyroscope data is approximately
364 microseconds, while the time left for processing
magnetometer data is approximately 344 microseconds.
Then the time for processing EMG signals is the time left
after processing IMU data. Figure 8 illustrates the results
from processing time measurement.
REFERENCES
[1]
B. Champaty, P. Dubey, S. Sahoo,S.S. Ray, K. Pal, A. Anis,

Development of Wireless EMG control system for rehabilitation
devices, Annual International Conference on Emerging Research
Areas: Magnetics, Machines and Drives (AICERA/iCMMD), July
2014
ADAPTIVE SILHOUETTE NORMALIZATION BY TRADITIONAL OF THE

DIFFERENTIAL OF NEIGHBORING SILHOUETTE
P. Nangtin, P. Kumhom, and K. Chamnongthai
Department of Electronic and Telecommunication engineering
King Mongkuts University of Technology Thonburi
Bangkok Thailand
E-mail : prasit2005@gmail.com, pkumhom@gmail.com, and kosin.cha@kmutt.ac.th
ABSTRACT
In this study, we focus on the silhouette normalization
problem in gait identify problem when he or she is occluded
at the same position and multiple times in the same video. We
propose an adaptive silhouette normalization by traditional
of the differential of neighboring silhouette. Firstly, we detect
a top and a bottom positions of the silhouette height and then
separate an occlusion and a non-occlusion periods. We
replace new position by an adaptive position method and
normalize silhouette to the standard silhouette. We adopt
Gait Energy Image (GEI) and extract their features by the
combined TDPCA and TDLDA, and identify gait by the
Euclidean distance. Finally, the EEPIT dataset with 135
classes and the CASIA datasets with 123 classes are used for
evaluating the proposed system. For the results, the
recognition rates of the proposed method indicate not only
effectiveness of separating the occlusion and the nonocclusion periods but also performance of adaptive silhouette
normalization.
separating a non-occlusion and an occlusion periods by the

differential of neighboring silhouette height as shown in Fig.
1 (2)-(a). Then they compensated the occluded position by
using average height of non-occlusion silhouette before and
after occlusion period as shown in Fig.1 (2)-(b). However, it
still 2 problems when the height difference frames varies less
than the differential threshold of neighboring silhouette
height.
Index Terms Silhouette normalization; Occlusion; Gait

identification;
1. INTRODUCTION
For previous researches related in human identification using
gait at a distance in partial occlusion, Han and Bhanu [1] used
the sum of silhouette image of the walking silhouette divided
by the number of silhouette images called Gait Energy Image
(GEI). It consists of static feature (major shapes of silhouettes
e.g., head, body) and dynamic feature (their changes over the
gait cycle e.g., limbs move forward and backward) so that it
can deal with some occlusions occurring in few frames in
sequences. Preprocessing before GEI construct is silhouette
normalization process. Han and Bhanu [1] use a top position
and a bottom positions of the silhouette image for referring
positions in resizing silhouette to standard silhouette.
An error is occurred when the reference position is
occluded in some frames as (2) shown in Fig. 1 (1)-(a) so
that some components of human body will be placed in
another position as (2) shown in Fig. 1 (1)-(b). Nangtin et
al. [2] used the differential threshold of neighboring height for
Fig. 1 Silhouette normalization problem

This paper is organized as follows. Section 2 presents basic
concept and Section 3 explained the proposed method.
Experiment set-up and results show in Section 4. Section 5 is
discussion. Finally, Section 6 concludes this paper.
2. BASIC CNOCEPT
Two problems occur when we use the differential of
neighboring silhouette height [2] for separate occlusion and
non-occlusions period silhouette normalization with
silhouette sequence. The first problem is the system cannot
separate occlusion and non-occlusion periods because each
frame has the differential of neighboring silhouette height
((1) and (2)) less than threshold ( ) as shown in Fig.
2 (a). The second problem is the system miss reference
position compensation because the average height includes
silhouette height in occlusion period ((3)) so that the result

as shown in Fig. 2 (b).
For our basic idea, we use traditional of the differential of
neighboring silhouette height ((1) + (2)). It can
separate occlusion period and non-occlusion period because
it ((1) + (2)) is more than threshold ( ). We use
adaptive average height of non-occlusion period (only one
period) and non-occluded reference position for
compensating the reference position as the silhouette
normalization result shown in Fig. 2 (2)-(b).
Firstly, we construct history threshold from training database

for deciding occlusion period. History threshold consists
height, top positon, and bottom position of silhouette. The
history threshold is calculated as follows.

= |( + 1) ()|, (1)
= 1 = 1 = 1
where () represents height, top position, and bottom
position of silhouette, height is distance between bottom
and top positions at frame , is number of class, is
number of template, is number of frame, is class, is
template, and is frame.
Then we separate occlusion and non-occlusion by
comparing history of height, top position, and bottom position
in testing silhouette image sequence with its threshold. If
history is more than threshold that frame is in
occlusion period. The history of silhouette height, the top
silhouette position, and the bottom silhouette position is
calculated as follows.
= 2
(2)
1(( + 1) ()),
where x1 is the first frame of changing height direction,
and x2 is the last frame of changing height direction.
Table 1 Adaptive reference position rule.
Rule
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
Fig. 2 Basic Concept

3. PROPOSED METHOD
In this paper, we focus on the silhouette normalization
problem in gait identify problem when something slightly
occludes the top or the bottom silhouette position. We
propose adaptive silhouette normalization. Firstly, we prepare
silhouette sequence by using method of Han and Bhanu [1].
Then we detect occlusion frame for separating occlusion and
non-occlusion periods. We adapt the reference position before
silhouette normalization. We evaluate the proposed method
by recognition rate of gait identification. We construct GEI
[1], extract feature by combined TDPCA and TDLDA [4],
and identify gait by Euclidean distance method. The details
are as following.
3.1 Separating occlusion and non-occlusion period
Ip1
1
1
1
1
1
1
1
1
1
0
Ip2
1
1
1
0
0
0
0
0
0
x
Input
Ip3
Ip4
0
1
0
0
0
0
1
1
1
0
1
0
0
1
0
0
0
0
x
x
Ip5
0
1
0
0
1
0
0
1
0
x
Ip6
0
0
1
0
0
1
0
0
1
x
Output
Op1
Op2
O11
0
O12
0
O13
0
0
O21
0
O22
0
O23
O14
O24
O15
O25
O16
O26
0
0
0: non-active; 1: active; x: dont care

3.2 Adaptive reference positions
We adapt reference positions, which are a top and bottom
positons, by R1 rule to R10 rule in Table 1. Adaptive
reference position consists 10 rules (R1 R10), 6 inputs (In1
- In6), and 2 outputs (Op1 Op2). The symbol in Table 1
explained in Table 2.
1
2 1 +1)
2
=
(),
1
(3)
where () is the height of silhouette frame and 1

and 2 are the first frame and the last frame in nonocclusion period which is after occlusion period,
respectively.
2
=
() ,
1
2 1 +1)
where () is the height of silhouette frame in nonocclusion period which is before occlusion period.
= (
1
2 1 +1)
= ( + ),
(5)
2
=
() ,
1
(6)
() is the top position of silhouette frame in

where
non-occlusion period which is after occlusion period.
Table 2 Meaning of symbols in Table 1.

Symbol
In1
In2
In3
In4
In5
In6
Op1
Op2
O11
O12
O13
O14
O15
O16
O21
O22
O23
O24
O25
O26
Meaning
Occlusion state of height of silhouette image
Occlusion state of top position of silhouette
image
Occlusion state of bottom position of silhouette
image
Occlusion period at first frame of silhouette
image sequence
Occlusion period at middle of silhouette image
sequence
Occlusion period at last frame of silhouette
image sequence
Compensating top reference position ()
Compensating bottom reference position ()
() = () +
() = () +
()
()
=
+
()
() =
()
=
()
= ()
()
= ()
() = () +
()
() =
()
= (
1
2 1 +1)
2
=
() ,
1
= (
(4)
(7)
()
= (
(8)
2
=
() ,
1
(9)
1
2 1 +1)
()
where
is the bottom position of silhouette frame
in non-occlusion period which is after occlusion period.
2
=
() ,
1
(10)
where
is the bottom position of silhouette frame
in non-occlusion period which is before occlusion period.
(
+
)
(11)
Finally, both adapted top and bottom position are referred

in silhouette normalization process.
3.3 Gait identification
We construct GEI by method of [1], extract GEI feature by
the combined TDPCA and LDLDA method [4], and identify
by Euclidean distance.
A. Feature Extraction
Transformation matrix of the combined TDPCA and
TDLDA method is calculated as follows.
= ,
(12)
where and is the largest eigenvector which represents

the feature matrix of TDPCA and TDLDA, respectively.
B. Training feature
For all training feature, given training set containing
templates from each class , feature , is computed by
projecting (, ) on to the feature space by the transformation
matrix following Eq. (13).
, = , ,
(13)
where is class and is template.

C. Testing feature
For testing feature, we generate testing feature ()following
Eq. (14)
= ,
(14)
where is the testing GEI template.

D. Gait identification
Finally, we adopt the Euclidean distance for measuring a
similarity between testing feature () and training feature
(, ) within class . We assign the testing GEI to class if

= , ,
= 1 = 1
() is the top position of silhouette frame in

where
non-occlusion period which is before occlusion period.
( + )
= 2 ,
1
2 1 +1)
(15)
where is number of class, is number of template,

is number of frame, is class, is template, and is frame.
4. EXPERIMENT SET-UP AND RESULTS
We perform 2 experiments from 2 datasets, the CASIA and
EEPIT datasets, to evaluate our approach. For EEPIT [5], it
consists of 135 people with side view sequence. We use three
normal sequences for training. For testing, we use two Double

upper Position (DuP) and two Double lower Position (DlP)
sequences for testing as shown in Fig. 3.
position is occluded, but also in frame which both the top and
the bottom positions are occluded. Moreover, we can use
information of GEI, which is constructed from the adaptive
reference position, in advancing gait identification as
separating the occluded GEI part and the non-occluded GEI
part as Nangtin et al. [2].
Table 3 Recognition rate.
Dataset
Sequence
EEPIT
DuP
DlP
LfUp
LfLp
MfUp
MfLp
RfUp
RfLp
CASIA
Normalization method
NonAdaptive
adaptive
66.67
69.67
20.00
26.53
33.60
40.52
56.91
59.85
23.85
30.53
48.24
53.78
32.52
35.24
67.48
69.52
Different
3.00
6.53
6.92
2.94
6.68
5.54
2.72
2.04
6. CONCLUSION
Fig. 3 An example of type of occlusion on all scenes in
EEPIT dataset [5].
This paper proposed the adaptive silhouette normalization

method for solving the silhouette normalization problem in
gait identify problem when he or she is occluded at the same
position and multiple times in the same video. We evaluate
with EEPIT dataset with 135 classes and CASIA dataset with
123 classes. The performance indicated not only effectiveness
of detecting the occlusion and the non-occlusion periods but
also significance of adaptive silhouette normalization method
over than non- silhouette normalization method [2].
7. REFERENCES
Fig. 4 An example of type of occlusion after superimposed a

black strip on all scenes in Casia dataset [6]
For CASIA [6], we select 123 subjects with side view
sequence. We use three normal sequences for training. For
occlusion testing, it consists 6 combinations; two Left frame
& Upper position (LfUp), two Left frame & Lower position
(LfLp), two Middle frame & Upper position (MfUp), two
Middle frame & Lower position (MfLp), two Right frame &
Upper position (RfUp), and two Right frame & Lower
position (RfLp) sequences as shown in Fig. 4.
Table 3 shows the effectiveness of the adaptive silhouette
normalization method over the non-adaptive normalization
method. For EEPIT dataset [5], they are 3 and 6.53 in DuP
and DlP sequences, respectively. For CASIA [6], they are
6.92, 2.94, 6.84, 5.54, 2.72, and 2.04 in LfPu, LfPl, MfPu,
MfPl, RfPu, and RfPl sequences, respectively.
5. DISCUSSIONS
From Table 3, not only the adaptive silhouette normalization
method improves in frame which the top or the bottom
[1] Ju Han and Bir Bhanu, 2006, Individual recognition

using gait energy image, IEEE Trans. Pattern Anal.
Mach. Intell., Vol. 28 (2), February, pp. 1505-1518.
[2] Prasit Nangtin, Pinit Kumhom, and Kosin
Chamnogthai, 2016, Gait identification with partial
occlusion using six modules and consideration of
occluded module exclusion, J. Vis. Commun. Image.
R., Vol. 36, pp. 107-121.
[3] J.Lu and E.Zhang, 2007, Gait recognition for human
identification based on ICA and fuzzy SVM through
multiple views fusion, Pattern Recognition Letters,
Vol. 28, No. 16, pp. 2401-2411.
[4] P. Sanguansat, W. Asdornwised, S. Jitrapunkul, and S.
Marukatat, 2006, Two-dimensional linear discriminant
analysis of principle component vectors for face
recognition, IEICE Trans. Inform. Syst. Vol. 7, pp.
2164-2170.
[5] http://172.16.1.41/prasit for EEPIT dataset, Electrical
Engineering Department, Pathumwan Institute of
Technology.
[6] http://www.cbsr.ia.ac.cn/english/gait%20databases.asp
for Casia dataset B, Center for Biometrics and Security
Research (CBSR), Institute of Automation, Chinese
Academy of Sciences (CASIA).
Automatic Comparison System for IC Units Positions

Pattanai Sirisombatampai and Werapon Chiracharit
Departmen
Department of Electronic and Telecommunication Engineering
King Mongkuts University of Technology Thonburi
126 Pracha Uthit Rd., Bang Mod, Thung Khru, Bangkok 10140,, Thailand
pattanai.tor@gmail.com
Abstract In manual inspection in the IC industry, shape and

color of all ICs are on the same panel and basically similar to
each other,
her, so that it is hard to visually verify its position
comparing with map file from the machine. This paper presents a
method to automatically count and verify IC remaining positions
which major ICs have been picked up and packaged. The
proposed method analyzes
alyzes a device position in IC panel based on
pattern matching technique. The experimental results show that
the proposed method can detect IC positions accurately
comparing to the map file.
KeywordsImage
Image processing, IC unit, Pattern Matching
technique
I. INTRODUCTION
Fig. 1 Environment setup

s
Machine vision technology is useful tools used in various

fields to reduce human error such as monitoring vegetation on
railway [1], stack coin counting [2], estimation of face
direction [3], or industrial applications, it has also been widely
used in process detection in recent years, such as HDD
inspection process [4], and counting residue wafer image [[5].
Pattern matching can be one the most important algorithm iin a
machine vision system. It is applied to detecting
etecting defect IC
image [6].
All of ICs in IC panel are tested
tested. After finishing,
Tester will generate the result called Map file.
Consequently, the passers are sorted to other process refer
map file and the failures are sent to human inspection,
comparing between position of ICs and map ffile to confirm
precisely machine. Sometimes machine sorted good devices, it
is error to pick up rejected device. Therefore, the purpose of
this study is to use the remaining device image to
automatically count the number and verify the correct IC
position.
III. PROPOSED
ROPOSE METHOD
II. ENVIROMENT SETUP

There is condition that needs to be controlled on this
experiment. The IC panel needs to be placed with white
surrounding background, So that pattern matching can be
processed effectively. It started IC panel has been complete
tested thenn machine will generate map file and keep on server
then take IC map panel image by scan into scanner keep the
image waiting proceed shown as Fig. 1.
Fig. 2 Proposed method diagram for image processing
A. IC panel detection
We propose a method to inspect the IC panel from the
input image in this topic which can divide in to three phases.
1) Convert RGB to HSL
The RGB image was contain image size 32 bits, so we need
to decrease image size to 8 bits. The method of HSL
conversion is the properly for clarify, this paper choose and
coverts RGB image to HSL.
2) Set reference pattern
After convert image to HSL, we need to set reference
pattern for start from this point to search panel edge in image,
this paper we propose to use ring notch as reference pattern.
3) Set panel edge
After get reference pattern, In this paper use edge
detection to set region of interest to find panel edge of 4 sides
by search from bright to dark as Fig. 3. After set panel edge
complete. We will get the region of interest and this function
we will use panel edge position for setting offset pixel to find
start position
Fig. 4 Pattern matching technique

For this function is required master template to compare
how many IC matching with this template. In this paper
setting the minimum score is 65% because this score can be
detect all IC units. After using pattern matching technique, we
will get number IC that matching as Fig. 5.
Fig. 5 Pattern matching technique

Fig. 3 Edge detection for IC panel
B. Pattern matching and counting IC
Pattern matching Technique as Fig. 4 used to find the
residue IC panel by using Normalized Cross Correction as
refers to (1).
L 1 k 1
C(i, j )
w( x, y) f ( x i, y
j)
(1)
x 0 y 0
Where i = 0,1,,M-1
j = 0,1,,N-1 and the summation is taken over the
region in the image where w and f overlap.
C. 1st comparison IC image with map file

Figure 6(b) contains bin result of each unit position which
opposites position from the left to the right. Number 1 means
good device will be picked up. The good device is not shown
on IC panel as Fig. 6(a). Other number mean the reject
device, they are not selected and remained on IC panel. In this
paper uses number 3 and 4 from Map file for IC position
comparison. The completion of pattern matching technique
contains pixel position of IC that match with Master template.
Then, query panel edge position data that mentioned on Set
panel edge section then we set offset for axis x and axis y
from panel edge to 1st position on top right as follow (2).
(2)
P ( x , y ) E x O x np x , E y O y mp y
Where P(x,y) is IC position n in along x and y direction. Ex
and Ey is edge pixel position on top left that get from edge
detection topic. Ox and Oy is offset pixel from edge to first
device on left hand side. px and py are center pixel of IC units
in along x and y.
and right images are operated images by compare with map

file.
TABLE I COMPARE RESULT IN CASE ACTUAL IC IMAGE MATCH WITH MAP FILE
(a)
(b)
Fig. 6 Map file data opposite position with IC panel

a) Sample IC panel. b) Map file.
IC
panel
Actual
reject
unit
Pattern
matching
Particle
Reject unit
in map file
Compared
IC position
result
80
80
80
Correct
17
17
17
Correct
11
11
11
Correct
14
12
14
Correct
42
41
42
Correct
After comparing the IC position with map file we will

overlay object at the correct IC position with map file as
shown in Fig. 7.
(a)
Fig. 7 IC detection with pattern matching
(b)
D. 2nd comparison IC image with map file

Regarding to pattern matching technique, It cannot detect
the IC which a particle on body as Fig. 8. We set the minimum
score of pattern matching technique 65% that means a particle
on body IC cannot be detected by usiing only pattern matching
technique. This paper has added a second method to detect
particle IC and then to take particle IC position and add yellow
dot to compare with map file on remaining IC.
(c)
(d)
(a)
(b)
Fig. 8 IC detect image (a) Pattern Matching technique detect
IC (b) Particle method detect IC.
IV. RESULT AND DISCUSSION
Table I shows the result of IC detection device with the
difference panel in case of actual IC number and position
matches with map file. Figure 9 shows the result from Table I.
The left hand sides of Fig. 9 are original image. The middle
images are graphic image which convert from number of map
file number 3 and 4 will be red dot and blue dot in map file
(e)
Fig. 9 Compare result in case actual IC image match with
map file.
Figure 10 shows the compare pictures of remaining IC
image and map file between actual IC number and position
mismatch with map file. The application is added red dot to
easier to see the mismatch position. The compared result
shows a capability in detecting IC device in difference panel
in case actual IC mismatch with map file shown in Table II.
Actual reject unit is number IC reject which remaining on

the panel from the image. Pattern matching and Particle means
number IC which these algorithms can detect. Reject unit in
map file mean no reject unit which show in the map file and
Compared IC position result mean the comparison result
between actual unit and map file.
From Table I and Table II, they show that the proposed
method can be compared IC units position effectively by
comparing result shown.
Considering the samples result for proposed method. The
IC panel images from Fig. 9 and Fig. 10. might not enough to
verify the proposed method. More IC panel images are desired
to earn more accurate of the proposed method.
TABLE II COMPARED RESULT IN CASE ACTUAL IC IMAGE MIS MATCH WITH MAP
FILE
IC
panel
Actual
reject
unit
Pattern
matching
Particle
Reject unit
in map file
Compared
IC position
result
10
Found
Mismatch
24
24
25
Found
Mismatch
Found
Mismatch
12
12
13
Found
Mismatch
10
Found
Mismatch
V. CONCLUSION
This paper presents an automatic method for compare the
object position in IC panel images. As shown in the result, the
proposed method successes to count and compare the object
positions with map file. The application accuracy is 100% in
the existing 10 IC panel images. However, this method also
can reduce human error and faster than human compare IC
unit positions
(f)
(g)
REFERENCES
[1]
(h)
[2]
[3]
[4]
(i)
[5]
[6]
(j)
Fig. 10 Comparison results in case actual IC image mismatch
with map file.
R. G. Nyberg *, N. K. Gupta , S. Yella*, M. S. Dougherty*S. Chen, B.,

Machine Vision for Condition Monitoring Vegetation on Railway
Embankments, Railway Condition Monitoring (RCM 2014), 6th IET
Conference on, pp 1-7, Sep 2014.
Komkrit S., Kosin Ch, Automatic Counting Method of Stacking Coins
Using Reflection, ICESIT Conference 2015, pp 65-68, Jun 2015.
Junko Hirata, Masakazu Morimoto, Kensaku Fujii, Estimating face
direction from low resolution images,, Automation Congress, 2008.
WAC 2008. World pp. 1-6, Oct 2008.
Varagul, J., and Chamniprasart, K.Development of Automated
Visual Inspection Machine for HDD Manufacturing process.
Proceedings of International Conference on Data Storage Technology
(DST-CON) Year 2010.
H. T. Chang, R. J. Pan Automatic counting of packaged wafer die
based on machine vision Information Security and Intelligence Control
(ISIC), 2012 International Conference on pp 274-277, Year 2012.
Chien-Chih Wang, Bernard C. Jiang, Jing-You Lin, and Chien-Cheng
Machine Vision-Based Defect Detection in IC Images Using the Partial
Information Correlation Coefficient IEEE Transactions on
Semiconductor Manufacturing, Vol. 26, No.3, pp 378-384, Aug 2013.R.
Nicole, Title of paper with only first word capitalized, J. Name Stand.
Abbrev., in press.
The 10th TESA Top Gun Rally: Smart Classroom

Phakphoom Boonyanant
School of Information Technology
Mae Fah Luang University
Chiang Rai, Thailand 57100
phakphoom.boo@mfu.ac.th
AbstractThis paper presents the smart classroom concept

used in TESA Top Gun Rally 2016. Based on embedded system
technology, the design system includes 5 platforms, i.e, Raspberry
Pi, Android, STM32L0 Nucleo, cloud web server and MATLAB.
The system has been proposed in the 10th TESA Top Gun Rally
2016 as a problem to be implemented in 5-day camp. Among 230
participants, 43 teams from 27 universities, the best team in
overall performance is Appendix from Kasetsart University.
KeywordsTESA, TGR2016, Smart Classroom, Embedded

System, Raspberry Pi, Android, STM32L0, MATLAB
I.
INTRODUCTION
TESA Top Gun Rally (TGR) is a well-known student

competition on embedded system development. The event has
been host by Thailand Embedded System Association
(TESA), the group of people from universities and private
companies who want to promote the embedded system
technology in Thailand. Starting in the year 2005, the
challenged problems to has been changed every year. For this
year, the 10th TGR2016 has been held in Mae Fah Luang
University, Chiang Rai, under the theme Smart Classroom.
The concepts of smart classroom have been considerably
proposed during the last decade together with the upcoming of
the new technologies such as Internet of Things (IOT)
technology, cloud computing and big data. During the year
2003 to 2009, the researches were focused on long distance
learning. Many successful experiments have been reported in
literatures [1]-[6]. While some research groups [7][8] were
interested on context aware and ambient intelligent. More
sophisticated systems including with embedded systems were
introduced for the last 5 years. In [9], the authors used RFID
for checking students attendance via Internet. The studies of
interactive classroom have been reported in [10]-[12]. Signals
from multiple sensors including infrared, camera and sound
were collected to analyze the students attention and feedback
to the teacher in real-time [10]. The paper [11] used camera
and image processing to detect number of head nod or shake
of students. This allows teacher to know how many students
understanding the lecture in class. The smartphone response
system based on Twitter has been presented in [12]. Their
successfully experimental results for a large classroom were
reported.
In Thailand, starting in the year 2012, the Thai government

had a policy to provide tablet computers for grade 1 students
throughout the country. But the project did not success from
many reasons such as low-efficiency tablet computers, poor
battery, no maintenance policy, etc. However, the idea of
using tablet computers in classroom is interesting for Thai
academic. In 2013, Rattana Auewittaya School, the private
school in Chiang Mai, started to deploy smart classroom in
grade 1 class by the supporting from Intel. On the other hand,
the Rangsit University began to set up smart classroom in the
university by providing Samsung tablet for the freshmen. In
the year 2014, Microsoft started launching the smart
classroom project in two middle schools: Suankularb
Wittayalai Nonthaburi School and Wora-racha Thinatdamatu
Witthaya School.
This paper presents another idea of smart classroom
implemented on embedded system. The idea is used as
problems to challenge undergraduate students to implement
and finish it in a 5-day camp of the TESA Top Gun Rally.
II.
SYSTEM CHALLENGED
As shown in Figure I, the smart classroom system

designed for the 10th Top Gun Rally is composed of 5
platforms connected together:
Raspberry Pi: A Raspberry Pi board plays a central
role in the system. It includes the following functions:
a)
Scanning and receiving data from a clicker via

Bluetooth Low Energy (BLE)
b) Recording the received data into the database

c)
Being a web server that contains the question and

answer. The teacher will choose the question and
choices using smart phone.
d) Being a gateway to submit data to cloud.

Android phone: The application on a smart phone that
allows a teacher to control contents and questions to
be displayed on a monitor. The mobile will connect to
the Raspberry Pi via WiFi connection.
STM32L0 Nucleo: STM32L0 Nucleo is used as a
clicker for a student. There are 4 buttons on the
clicker. Once it is pressed, the data which is the
answer from a student will be sent and stored in the
Raspberry Pi before transferring to cloud. The

STM32L0 will be connected to the Raspberry Pi via
Bluetooth Low Energy (BLE).
Cloud web server. The web server is used to store data
from Raspberry Pi.
The problems have been assigned to the participants in

each day. The solutions obtained from each team would be
evaluated by the committee from TESA. The scores from all
teams participated in the TGR2016 are shown in Figure II in
histogram. The average score is at about 30.56.
MATLAB: A platform optimized for engineering and

scientific problem, is used to find the answer for the
question of ranking a given classroom compared to
overall. The data set is simulated for 100 classrooms,
10 questions and 20 students. The correct answers and
arrival time are generated by Poisson distribution. Kmeans clustering is applied to classify groups of data.
Fig. 2. Distribution of the score in TGR2016
IV. SUMMARY
Fig. 1.
System design for smart classrom in TGR2016
III. THE 10TH TOP GUN RALLY

The 10th TGR has been arranged during the January 1823, 2016 by TESA and the Mae Fah Luang University (MFU)
as a host in Chiang Rai, Thailand. There are 230 participants
in 43 teams from 27 universities. The event arranged as a 5day camp, 2-day for study the development tools and
techniques and the other 3-day for the project implementation.
The summary of the activity in each day is given in the Table I
TABLE I.
The paper presents the system architecture of smart

classroom which is the challenged problem in TGR2016. The
system is composed of 5 platforms connected together. The
participants must implement the system within 5 days. The
scores have been collected everyday till the last day. In this
year, the first prize was given to the team Appendix from
Kasetsart University.
To this end, according to the objective of the event, i.e., to
strengthen the embedded system technology in Thailand, the
TGR2016 has successfully been arranged by the collaboration
from many parts such as Mae Fah Luang University,
Technical committee from TESA, universities around the
country and sponsors from organizations and private sector.
References
5-DAY ACTIVITIES IN THE TGR2016
Date
18/1/2016
Activity
Lecture
Work shop
19/1/2016
20/1/2016
Lecture
Work shop
Assignment
21/1/2016
Assignment
22/1/2016
Assignment
Topics
MATLAB GUI & brief intro to
stat & machine learning toolbox
(distribution, cluster analysis)
STM32 programming and BLE
stack
local classroom computer
clicker device -> gateway +
screen <- smartphone
data synchronization
gateway database -> cloud server
analysis platform
MATLAB UI + cloud interface +
analysis
[1]
[2]
[3]
[4]
[5]
Y. Shi, W. Xie, G. Xu, R. Shi, E. Chen, Y. Mao, and F. Liu, The smart
classroom: merging technologies for seamless tele-education, IEEE
Pervasive Computing, Vol.2, pp. 47-55, April-June 2003.
C. Snow, J. M. Pullen, and P. McAndrews, Network EducationWare:
an open-source web-based system for synchronous distance education,
IEEE Transactions on Education, Vol. 48, pp. 705-712, Nov. 2005.
D. Pishva, Smart Classrooms Bring Top-Quality Education around the
Globe, in Proc. Int. Symp. On Applications and the Internet
Workshops, 2007, pp. 40-44, Jan 2007.
C. Di, Z. Gang, and X. Juhong, An Introduction to The Technology Of
Blending-reality Smart Classroom, in Proc. Int. Symp. On Knowledge
Acquisition and Modeling, 2008, pp. 516 519, 21-22 Dec. 2008.
G. G. D. Nishantha, D. Pishva, and Y. Hayashida, Smart Classrooms:
Architectural Requirements and Deployment Issues, in Proc. 2008
IEEE Region 10 and the Third int. Conf. on Industrial and Information
Systems, pp. 1-6, 8-10 Dec. 2008.
[6]
[7]
[8]
[9]
Y. Suo, N. Miyata, H. Morikawa, T. Ishida, and Y. Shi, Open Smart

Classroom: Extensible and Scalable Learning System in Smart Space
Using Web Service Technology, IEEE Transactions on Knowledge and
Data Engineering, Vol. 21, no. 6, pp. 814 828, Jun. 2009.
K. Scott and R. Benlamri, Context-Aware Services for Smart Learning
Spaces, IEEE Transactions on Learning Technologies, Vol. 3, no. 3,
Jul.-Sep. 2010.
X. Li, L. Feng, L. Zhou, and Y. Shi, Learning in an Ambient Intelligent
World: Enabling Technologies and Practices, IEEE Transactions on
Knowledge, Vol. 21, no. 6, pp. 910 924, Jun. 2009.
C. H. Chang, Smart Classroom Roll Caller System with IOT
Architecture, in Proc. 2011 Second Int. Conf. on Innovations in Bioinspired Computing and Applications, pp. 356 360, 16-18 Dec. 2011.
[10] N. Gligori, A. Uzelac, and S. Krco, Smart Classroom: Real-Time

Feedback on Lecture Quality, in Proc. 2012 IEEE Int. Conf. on
Pervasive Computing and Communications Workshops, pp. 391 394,
19-23 Mar. 2012.
[11] Y.-C. Yu, S. D. You, and D.-R. Tsai, Social Interaction Feedback
System for the Smart Classroom, in Proc. 2012 IEEE International
Conference on Consumer Electronics, pp. 500-501, 13-16 Jan. 2012.
[12] Y. Kim, S. Jeong, Y. Ji, S. Lee, K. H. Kwon, and J. W. Jeon,
Smartphone Response System Using Twitter to Enable Effective
Interaction and Improve Engagement in Large Classrooms, IEEE
Transactions on Education, Vol. 58, no. 2, pp. 98-103, May 2015.
Adaptive Hybrid Encryption in

Resource-and-Bandwidth-Constrained Systems with
Secret Mixing
Apirath Limmanee, Sorakrai Kraipui , Wiroon Sriborrirux, Nipon Sommai, and Soontree Jaikhong
Burapha University
ESTITA, Department of Electrical Engineering, Faculty of Engineering
Chonburi, Thailand
Email: apirath@eng.buu.ac.th, p.sorakrai@gmail.com, wiroon@eng.buu.ac.th
niponsommai@gmail.com, soontree.11ee@gmail.com
AbstractThe main objective of this paper is a general

formulation of an adaptive cryptosystem based on Maurers
idea of constructive cryptography. Our adaptive framework is
constructed in the context of hybrid encryption adapted to three
varying factors, which are the maximum data flow between the
transmit-receive parties, the computational complexity, and the
level of security required for each of them. Based on these factors,
the adaptive system will select the most appropriate encryption
scheme from four proposed models, which are traditional hybrid
encryption, hybrid encryption with secret mixing of key, hybrid
encryption with secret mixing of message, and hybrid encryption
with secret mixing of both key and message. We explain the four
models in detail, giving a comparison and analysis together with
some implementation results as a guideline on how our adaptive
system can make its choices.
I. I NTRODUCTION
When talking about a cryptosystem, Alice, Bob, and Eve are
often mentioned as standard parties involved. In general, Alice
and Bob want to communicate with each other in secret, while
Eve wants to know what they are talking about. Traditionally,
Alice and Bob have no conflicting goal. They usually agree
on a specific encryption-decryption scheme. However, with
the advance in communication technology nowadays, such as
the Internet-of-Things (IoT) technology, Alice and Bob are
not just two computers anymore, they can be anything from
smartphones and tablets to wearable sensors or home washing
machines. Therefore, although Alice and Bob have the same
goal of communicating securely, they may be in conflict once
we talk about the data rate at which each party would like
to transmit and receive data, as well as the computational
complexity each party is willing to spend for encryptiondecryption processes.
If we make an assumption that each of Alice and Bob has
in his or her disposal more than one encryption/decryption
schemes, their choice of scheme can be adapted to the resources they have at the time secret messages are communicated, such as the transmission bandwidth and the CPU
resource.
To help illustrating the adaptive framework, we describe a
specific system that we have developed as a concrete example,
the standard of our adaptation, as well as how it can be made

adaptive, in Section II. The system is called OBSeS (Open Burapha Secure e-Book Streamer) developed for a secure e-Book
streaming application of SE-EDUCATION Public Company
Limited. To emphasize the top-down approach, we dedicate
Section III for an explanation of the concept of specification
in constructive cryptography. Then, after briefly reviewing the
presence of hybrid encryption and secret mixing in literature
in Section IV, we will explain in detail the algorithms and
components of the four hybrid encryption models we have for
adaptation in Section V. After that, Section VI analyzes the
four models, providing comparisons in terms of the level of
security, the computational complexity and the required data
rate. This analysis, together with some implementation results,
allow us to make the system properly adaptive. The last section
concludes the work, relating the results and analysis to the
constructive cryptography framework discussed earlier.
II. S YSTEM A RCHITECTURE E XAMPLE AND I TS P OSSIBLE
A DAPTATION
Our online e-book streaming system is designed such that
the books can be read from the internet without being downloaded into the storage. In addition, digital rights management
(DRM) is included in the system to protect the intellectual
rights. When an e-book reader requests for a book, the e-book
reader application will make a connection to the application
server. Then, the server will take the book contents from ebook storage page-by-page, as shown in Fig. 1. Each page is
encrypted by the hybrid encryption technique, which combines
a symmetric-key algorithm with an asymmetric one for security improvement. The encrypted content is then streamed to
the e-book reader application, who performs decryption and
displays the page content. Since no content is stored in the
reader, this system reduces the required memory and improves
security at the same time.
In practice, there can be three possible types of variation
affecting the system performance. First, the transmission bandwidth may vary from time to time. Second, we can see from
the picture that the user can either be the smartphone or
Application Server
ebook's
page no.
R : A1 A2 ... An ! ,
Hybrid AES-ECC
Encryption
ebooks storage
Web Service
Ebook Reader
Content Streaming
Smartphone
Fig. 1.
Definition 1: An n-choice setting is defined as the relation
Laptop
System Architecture of OBSeS
the laptop, of which computational complexity differs. Third,

each part of the transmitted message may be of different
importance. For example, the header part should be better
protected than the content. The enemy cryptanalyst will not
be able to read the content unless he or she knows the header.
Therefore, in order to effectively control the system performance, our encryption should be adaptive to these three
factors in a top-down manner. By top-down manner, we mean
that once the three factors are learned from the environment,
they become the specifications at the abstract level to which
the encryption must be adapted. The lower, concrete level
then makes some available choices such that the abstract
specifications are satisfied.
where Ai is called the ith choice domain and is called the

effect space [1].
In a general adaptive cryptosystem where Alice and Bob
choose the appropriate encryption-decryption scheme, A1 and
A2 may refer to Alices and Bobs choices, respectively.
Although the choice of Eve the cryptanalyst might be added
as A3 , it might be impossible to learn about such a choice as
Eve will never cooperate. Therefore, we do not take it into
account.
In our context, as an example, we may have specifications
in two levels. In the abstract level, the choices of data rate,
computational complexity, and security level may lead to the
effect of having low or high delay, as well as successful or
failed communication. As for the specifications in the concrete
level, since this level is in the service of the abstract level, the
effect space of the concrete level will be the same as the choice
space Ai of the abstract level. Based on Definition 1, we can
formally describe the specification at the abstract level and at
the concrete level as follows.
Formulation 1: Example of specification formulation at the
abstract level
RA : AA1 AA2 ! A ,
(2)
where RA is the 2-choice setting at the abstract level, AA1
is the set of Alices choices, AA2 is the set of Bobs choices
and A is the resultant effect space. In this case, the choice
is selected from the following domain
AA1 = AA2
{(HD , HC , HS ), (HD , HC , LS ),
(HD , LC , HS ), (HD , LC , LS ),
(LD , HC , HS ), (LD , HC , LS ),
(LD , LC , HS ), (LD , LC , LS )},
III. A PPLICATION OF THE S PECIFICATION C ONCEPT IN

C ONSTRUCTIVE C RYPTOGRAPHY
Constructive cryptography proposed by Maurer makes use
of abstraction to formulate cryptosystems, allowing us to
take a top-down approach in the system design [1], [2]. The
framework defines and proves that a system satisfying an
abstract or ideal specification is realized by systems satisfying
concrete or real specifications. The work in [2] discusses in
detail how to construct a complex cryptosystem from simpler
components, which are further derived from even simpler
ones. Thus, the key concepts of constructive cryptography are
the concepts of specification as well as the reduction or
step-wise refinement. In this work, we focus on the former
concept.
The constructive cryptographys approach describes a general setting where n choices are made by n parties who do
not necessarily collaborate. The n-tuple of choices result in
an effect belonging to a set called the effect space, as defined
below.
(1)
(3)
where HD and LD means high and low data rate, HC and

LC means high and low computational complexity, HS and
LS means high and low security level, respectively. The choice
results in the effect space
A
{(!HD , !SC ), (!HD , !F C ),

(!LD , !SC ), (!LD , !F C )},
(4)
where !HD and !LD means high and low delay, !SC and
!F C means success communication and failed communication, respectively.
Formulation 2: Example of specification formulation at the
concrete level
RC : AC1 AC2 ! C ,
(5)
where RC is the 2-choice setting at the concrete level, AC1

is the set of Alices choices, AC2 is the set of Bobs choices
and C is the resultant effect space. Since the concrete level
is in the service of the abstract level, the effect space of the
concrete level will be the same as the choice space at the

abstract level.
C = AA1 = AA2
(6)
mixing takes a vital role again as a component in the set of

four encryption models from which our adaptive system makes
its choices.
The instances in the choice domain AA1 and AA2 , as well

as in the effect space A and C shown in Formulation 1
and 2 are represented by some notations indicating the high
or low levels of important factors. In practice, of course, the
instances should be represented by some quantity specifying
more than just being high or low.
The choices AC1 and AC2 in the concrete level will be taken
from the four hybrid encryption models discussed in Section
V. As a prerequisite for understanding the four models, we
discuss the general idea of hybrid encryption and secret mixing
in the next section.
B. Secret Mixing
IV. H YBRID E NCRYPTION A ND S ECRET M IXING : A B RIEF

R EVIEW
A. Hybrid Encryption
The word hybrid in hybrid encryption implies that two
or more encryption schemes are combined together. Although
some papers, such as [3], propose a combination of two
symmetric-key encryption schemes in order to enhance the
performance, most hybrid encryption schemes proposed in
literature, including the seminal papers from Cramer and
Shoup [8], [9], define hybrid encryption as a scheme employing an asymmetric encryption technique to encrypt a
symmetric key, which is then used later for encrypting the
secret message. Therefore, a hybrid scheme consists of two
main parts, the key encapsulation mechanism (KEM) and the
symmetric-key encryption. In this work, we use elliptic curve
cryptography (ECC) for the KEM part to protect the AES
(Advance Encryption Standard) key used for symmetric-key
encryption.
KEM consists of three phases, which are public-key generation, encryption, and decryption, involving two legitimate
parties usually named Alice and Bob. In the first phase, Alice
randomly selects a number from a group of order q and
uses this number to generate a private key and a public key.
Alice keeps the former secret and transmits the latter to Bob.
In the encryption phase, Bob randomly generates a secret
symmetric key and encrypt it into a ciphertext by using an
asymmetric algorithm with the public key from Alice. The
resulting ciphertext is transmitted to Alice. In the last phase,
Alice uses the private key to decrypt the ciphertext and obtain
Bobs symmetric key.
In the second part of hybrid encryption, Alice and Bob can
simply use the shared symmetric key derived from the KEM
part to encrypt their secret messages.
There are several variations of hybrid encryption in earlier
works, such as [4], [8], [9], as well as in more recent works,
such as [5], [6], [7], [13]. They usually vary in the choices of
symmetric key and public key encryption.
The secret mixing technique has already been suggested in
our earlier works on hybrid encryption as well as unequal
security protection (USP) [14], [15]. In this work, secret
The secret mixing problem discussed in this paper is a

special case of weakly secure network coding discussed in
[12]. Introduced by Ahlswede et al. in 2000 [10], the works in
network coding used for multicasting has led to an information
mixing technique employed as a means to achieve security, the
technique called secure network coding [11].
Some authors of this work have published the use of secret
mixing in [14], [15]. In [14], secret mixing used is a simple
XOR operation employed in the message encryption to reduce
the overall encryption and decryption time. In [15], we use
secret mixing to mix together several parts of AES key in
hybrid encryption. Then each mixture will be protected by
different ECC keys such that security is enhanced.
Though the detailed application of secret mixing in our work
will be discussed in following sections, a simple example can
be given for illustration. Suppose that we have two secret
symbols b1 and b2 . What is more secure between 1) encrypting
each of them separately before transmission or 2) transmitting
the encrypted version of the mixtures b1 + b2 and b1 + 2b2 ?
The correct answer is 2) because if the enemy can break
one of two ciphers, the method 1) will expose half of the
data to the enemy, whereas in the mixing method, the exact
knowledge of both b1 and b2 is still protected. Note that the
legitimate receiver can perform decryption to derive b1 + b2
and b1 + 2b2 and easily solve for b1 = 2(b1 + b2 ) (b1 + 2b2 )
and b2 = (b1 + 2b2 ) (b1 + b2 ).
V. T HE F OUR M ODELS OF H YBRID E NCRYPTION
S CHEMES U SED FOR A DAPTATION : T HEIR A LGORITHMS
AND P ROTOCOLS
The previous section discusses the general ideas of hybrid
encryption and secret mixing. This section will scrutinize
them more in detail regarding their basic components and
algorithms.
As earlier discussed, traditional hybrid encryption can be
broken into two components, which are KEM and the symmetric key encryption. However, our adaptive system consists
not only of the traditional hybrid encryption, but also three
other models. We can differentiate the four proposed models
by looking at how they differ in the KEM and the symmetric
key encryption parts, as shown in the table below.
From the table, we can see that we can explain the four
models by describing four basic components, which are normal
KEM, KEM with secret mixing, normal AES, and AES with
secret mixing. We do so in the following subsections.
A. Normal Key Encapsulation Mechanism (KEM)
With the assumption that Bob would like to transmit a secret
message to Alice, KEM consists of three sub-processes, Alice
and Bobs public key generation, Bobs key encapsulation, and
Alices key decapsulation. The basic idea is that both Alice
TABLE I
D IFFERENCES IN THE BASIC C OMPONENTS OF THE F OUR H YBRID
E NCRYPTION M ODELS U SED FOR A DAPTATION
3) Key Decapsulation of Alice:

3.1) Alice derives the shared secret x using R from Bob
and her private ECC key dA .
Model
KEM
Symmetric Key
Encryption
1. Traditional Hyb. Enc. (Hyb)
Normal KEM (A)
Normal AES (C)
2. Hyb. Enc. with Secr. Mix.

of Key (HybMixK)
KEM with
Secr. Mix. (B)
Normal AES (C)
3. Hyb. Enc. with Secr. Mix.

of Message (HybMixM)
Normal KEM (A)
AES with
Secr. Mix. (D)
4. Hyb. Enc. with Secr. Mix. of

Key and Message (HybMixKM)
KEM with
Secr. Mix. (B)
AES with
Secr. Mix. (D)
(x, y) = RdA
(13)
3.2) Using the shared secret x and the received ciphertext

CK as input, Alice performs the elliptic-curve decryption to
derive the key KAES .
KAES = DK (x, CK )
(14)
and Bob can generate the same shared secret by using, in

Alices case, her private key together with Bobs public key,
and in Bobs case, his private key together with Alices public
one. The step-by-step algorithm is summarized below.
1) Public Key Generation of Alice and Bob:
1.1) Alice and Bob independently and randomly generate
her/his own private key dA and dB , respectively, from a group
of order q.
dA
dB
Zq ,
Zq ,
(7)
1.2) Based on dA and dB , Alice and Bob generate her/his

public key QA and QB , respectively, from generator G.
QA
dA G,
(9)
QB
dB G,
(10)
1.3) Alice and Bob send her/his public key (Rs PK#1,
Ss PK#1) to each other as shown in Fig. 2. Only Alices
public key will be used in KEM. Bobs public key will be
used for creating the digital signature discussed in the system
implementation part.
2) Key Encapsulation of Bob:
2.1) Bob generates the shared secret x using a random
number r and Alices public key QA .
(x, y) = rQA = rGdA = RdA
B. Key Encapsulation Mechanism (KEM) With Secret-Mixing

With the proposed secret mixing, there are some differences
from the previous case, which are summarized below.
1) Public Key Generation of Alice and Bob:
1.1) Alice and Bob independently and randomly generate n
private keys dA,i and dB,i , respectively, where i = 1, 2, ..., n,
from a group of order q.
dA,i
dB,i
Zq ,
Zq ,
(15)
(16)
1.2) Based on each of dA,i and dB,i , Alice and Bob generate
her/his public key QA,i and QB,i , respectively, from the
generator G. Only Alices public keys will be used in KEM.
Bobs public keys will be used for creating the digital signature
discussed in the system implementation part.
(12)
The key KAES is kept secret by Bob, whereas CK , together

with R in (5), is transmitted to Alice as the AES Key
Cipherblock #1 according to Fig. 2.
Normal Key Encapsulation Mechanism
As can be seen in Fig. 2, steps 1)-3) of KEM without secret

mixing are repeated periodically such that Alice can use the
AES Key Cipherblock #i, i = 2, 3, ..., N , to derive the ith
update of the symmetric key KAES . The regular key updating
leads to better security.
(11)
2.2) Bob randomly generates an AES key KAES .

2.3) Employing the elliptic-curve encryption EK with the
derived shared secret x and the AES key KAES as input, Bob
generates the ciphertext CK encapsulating the key.
CK = EK (x, KAES )
Fig. 2.
(8)
QA,i
dA,i G,
(17)
QB,i
dB,i G,
(18)
1.3) Alice and Bob send her/his first public keys QA,1 (Rs
PK#11)and QB,1 (Ss PK#11) to each other as shown in Fig. 3.
3) Key Decapsulation of Alice:

3.1) Alice derives all the shared secrets xi , i = 1, 2, ..., n,
using Ri from Bob and her private ECC keys dA,i .
(xi , yi ) = Ri dA,i
(23)
3.2) Using the shared secret xi and the received ciphertext

cK,i as input, Alice performs the elliptic-curve decryption to
derive the mixed secret mi .
mi = DK (xi , cK,i )
Fig. 3.
Key Encapsulation Mechanism with Secret Mixing
2) Key Encapsulation of Bob:

2.1) Bob generates the shared secrets xi , i = 1, 2, ..., n
using randomly generated numbers ri and Alices public key
QA,i .
(xi , yi ) = ri QA,i = ri GdA,i = Ri dA,i
(19)
2.2) Bob randomly generates an AES key KAES
2.3) The AES key is partitioned into n sections with the
equal number of bits.
KAES = [KAES,1 , KAES,2 , ..., KAES,n ]
(20)
2.4) All sections of the AES key are mixed (linearly

combined) together into n mixed secrets mi , i = 1, 2, ..., n
as follows.
mi = ai,1 KAES,1 + ai,2 KAES,2 + ... + ai,n KAES,n (21)
where all aij 2 {0, 1, 2, ..., (q 1)}, j = 1, 2, ..., n,
must be known to both parties.
2.5) Employing the elliptic-curve encryption EK with each
derived shared secrets xi and each mixed secret mi of the
AES key, i = 1, 2, ..., n, as input, Bob generates ciphertexts
cK,i encapsulating the AES key.
cK,i = EK (xi , mi )
(22)
The mixed key mi is kept secret by Bob, whereas cK,i is

transmitted to Alice as the AES Key Cipherblock # 1i in the
order shown in Fig. 3. Note that the cipherblock request from
Alice (REQ AES Key Cipherblock # 1i) and the key exchange
(Ss PK # 1i, Rs PK # 1i) must be performed earlier.
(24)
3.3) With the mi derived from 3.2) and the known parameters aj,k ,j = 1, 2, ..., n, k = 1, 2, ..., n, the system of n
equations in (21) is solved to find KAES,i , i = 1, 2, ..., n.
3.4) All parts KAES,i , i = 1, 2, ..., n, are assembled into
the full AES key KAES .
As can be seen in Fig. 3, steps 1)-3) of KEM with secret
mixing are repeated periodically such that Alice can use the
AES Key Cipherblock # ji, j = 1, 2, ..., N , i = 1, 2, ..., n to
derive the ith partition of the j th updates of the symmetric
key KAES . The regular key updating leads to better security.
In our system, if the AES key is partitioned into four parts
before mixing, i.e., n = 4 in Eqs. (20) and (21). According to
[15], we can express the secret mixing in the following matrix
equation.
2
3 2
3 2
3
m1
1 1 1 1
KAES,1
6 m2 7 6 1 2 3 4 7 6 KAES,2 7
6
7 6
7 6
7
(25)
4 m3 5 = 4 1 3 6 8 5 4 KAES,3 5
m4
1 5 8 15
KAES,4
Note, however, that the proposed mixture is not the only
secret mixing possible. Any mixing, with the property that
the missing knowledge of any mixture left the enemy clueless
about every key partition, can be considered optimal.
C. Normal AES Symmetric Key Encryption
After Bob finishes key encapsulation in the KEM part, either
with or without secret mixing, he uses the encapsulated AES
key for symmetric key encryption of his plaintext.
D. AES Symmetric Key Encryption with Secret Mixing
The difference between normal AES encryption and AES
encryption with secret mixing can be illustrated in Fig. 4.
Unlike the normal AES encryption shown in Fig. 4(a), AES
encryption with secret mixing divides the secret message into
two parts, as shown in Fig. 4(b). Each symbol of the first
part, denoted by x1 , is encrypted by AES, whereas that of
the second part, denoted by x2 , is XORed with x1 bit by bit.
The encrypted version of x1 , represented by A(x1 ), and the
mixed information x1 x2 are transmitted to the receiver, who
derives x1 by decrypting the AES cipher and derives x2 by
subtracting x1 from the received x1 x2 .
Since this system encrypts only a half message with AES,
we can save the encryption and decryption time by about one
half of that in pure AES.
Note that, when compared with normal AES, the amount
of security decrease in AES with secret mixing is negligible.
When the enemy listens to the unencrypted part x1 x2 , he

still does not know either x1 or x2 unless he breaks the AES
cipher first to recover x1 . And even if he only has to break
half of the cipher, he will still has to find some ways to hack,
compute, or guess for the whole AES key like he has to do
in a normal AES. However, as the enemy is given an extra
knowledge of x1 x2 , this type of security is termed by [12]
as weak security.
Obviously, at the expense of higher security, secret mixing

of key requires that the system sends more data in the KEM
process, thus requiring more bandwidth.
Fig. 5. Test Result for the Case of 4 Key Partitions (n = 4) without Secret
Mixing (a) and with Secret Mixing (b)
Fig. 4. Comparison between (a) Normal AES Encryption and (b) AES
Encryption with Secret Mixing
VI. A NALYSIS OF THE F OUR M ODELS IN T ERMS OF

S ECURITY L EVEL , C OMPUTATIONAL C OMPLEXITY, AND
DATA R ATE
The previous section has already shown the components of
four models of hybrid encryption. This section will analyze
and compare them in terms of security level, computational
complexity, and data rate.
TABLE II
C OMPARISON OF THE F OUR H YBRID E NCRYPTION M ODELS IN T HREE
A SPECTS
Aspects
Comparison
1. Security Level
2. Comput. Comp.
3. Data Rate
HybMixK HybMixKM > Hyb HybMixM

HybMixM HybMixKM < Hyb HybMixK
Hyb = HybMixM < HybMixK = HybMixKM
With the mixing of AES key parts and protecting the

mixture by different ECC keys, as discussed in V-B, we
significantly increase the security level since the enemy has
to attack n = 4 ECC keys instead of just one. This is why,
as shown in Table II, HybMixK and HybMixKM are more
secure than Hyb and HybMixM.
As discussed in V-D, when the secret message is mixed,
the level of security drops only by a negligible amount as
compared with the non-mixing case. That is why we say
HybMixK HybMixKM and Hyb HybMixM in terms of
security.
According to V-D secret mixing of message reduces the
computational complexity needed for encryption and decryption by a half. As for the secret mixing of key, the result in
Fig. 5, derived from the implementation in the SE-ED Cloud
virtual machine based on Intel(R) Xeon(R) CPU E7-4870 at
2.40GHz with 4GB RAM under 64-bit Windows Web Server
2008 R2, shows that, for n = 4, the increase on CPU load due
to secret mixing is negligible. Thus, in terms of computational
complexity, HybMixM HybMixKM and Hyb HybMixK.
VII. C ONCLUSION AND S UGGESTION FOR F UTURE W ORK

From the previous sections analysis, we can relate the
properties of the four models to the constructive framework
as follows. First, Alice and Bob has to make specifications
at the abstract level as AA1 and AA2 in Formulation 1. The
specifications will be in the forms of how high the data rate,
the computational complexity, and the security level have to
be in order to achieve a successful and secure communication.
After that, Alice and Bob has to reach a mutual agreement
regarding the hybrid encryption model that will used, i.e.,
AC1 = AC2 and will be chosen from the set {Hyb, HybMixK,
HybMixM, HybMixKM}.
For a more practical implementation, AA1 and AA2 could
be quantitatively determined, e.g. by the number of megabytes
per seconds indicating the required data rate, by the encryption/decryption time in microseconds, and by the time required
to break the cipher. In addition, we may allow the choices AC1
and AC2 to be selected from a larger set. For example, not
only the models, but also the key size used in the models and
the rate of AES key updating may be chosen from a finite
pool.
R EFERENCES
[1] U. Maurer and R. Renner, Abstract Cryptography, Innovations in
Computer Science, 2011.
[2] U. Maurer, A. Ruedlinger, and B. Tackmann, Confidentiality and Integrity: A Constructive Perspective, Theory of Cryptography, pp. 209229, Springer Berlin Heidelberg, 2012.
[3] M.B. Vishnu, S.K. Tiong., M. Zaini, and S.P. Koh, Security Enhancement of Digital Motion Image Transmission Using Hybrid AES-DES
Algorithm, Asia-Pacific Conf. on Communications, Tokyo, Oct. 2008.
[4] M. T. El-Hadidi, N. H. Hegazi and H. K. Aslan, Implementation of a Hybrid Encryption Scheme for Ethernet Computers and Communications,
pp. 150-156, IEEE Symposium, 1995.
[5] M. Ren, and Z. Miao, A Hybrid Encryption Algorithm Based on
DES and RSA in Bluetooth Communication Modeling, Simulation and
Visualization Methods (WMSVM), pp. 221-225, IEEE, 2010.
[6] X. Hu, and L. Ma, A Study on the hybrid encryption technology in the
security transmission of electronic documents Information Science and
Management Engineering (ISME), pp. 60-63, IEEE, 2010.
[7] Y. Alkady, M. I. Habib, R Y. Rizk, A New Security Protocol Using
Hybrid Cryptography Algorithms Computer Engineering Conference
(ICENCO), pp. 109-115, IEEE, 2013.
[8] R. Cramer and V. Shoup, A Practical Public Key Cryptosystem Provably Secure against Adaptive Chosen Ciphertext Attack, Advances in
Cryptology-CRYPTO98, pp. 13-25, Springer, 1998.
[9] R. Cramer and V. Shoup, Design and analysis of practical public-key
encryption schemes secure against adaptive chosen ciphertext attack,
SIAM Journal on Computing, vol. 33, no. 1, pp. 167-226, 2003.
[10] R. Ahlswede, N. Cai, S.-Y.R. Li, and R.W. Yeung, Network Information
Flow, IEEE Trans. Inform. Theory, vol. 46, no. 4, pp. 1204-1216, Jul.
2000.
[11] N. Cai and R.W. Yeung, Secure Network Coding, Int. Symp. Information Theory, Jun. 2002.
[12] K. Bhattad and K.R. Narayanan, Weakly Secure Network Coding, in
Proc. NETCOD, Apr. 2005.
[13] S.-H. Zhu, Research of Hybrid Cipher Algorithm Application to
Hydraulic Information Transmission 2011 Int. Conf. on Electronics,
Communications and Control (ICECC), Ningbo, Sep. 2011.
[14] W. Sriborrirux, A. Limmanee, and S. Kraipui, Unequal Security
Protection: A Unified Framework, Implementation, and Performance
Evaluation of Theoretical and Practical Security Int. Symp. on Wireless
Communication Systems (ISWCS), Ilmenau, Aug. 2013.
[15] A. Limmanee, W. Sriborrirux, and S. Kraipui, Hybrid Encryption
Scheme for Digital Content with Key Partitioning and Secret Mixing:
Design and Implementation Int. Symp. on Wireless Communication
Systems (ISWCS), Ilmenau, Aug. 2013.
Circuit Analysis for Heartbeat Measurement

Anuparp Boonsongsrikul
Nakrom Massaya
Jirayut Phangklin
Dept. of Electrical Engineering,

Email: anuparp@eng.buu.ac.th

Email: 54550961@go.buu.ac.th

Email: 54550613@go.buu.ac.th
AbstractHealth monitoring is a key performance indicator

reflecting quality of life. Previous circuit and signal analysis
work proposed a variety of measurement techniques but they
incompletely explain in terms of analog and discrete signal
analysis. This paper aims to study continue analog and discrete
signals and fulfil explanation of circuit analysis for heartbeat
measurement. The analyzed circuit consists of a heartbeat sensor,
a noise filter, a signal amplifier and digital signal processing. Our
contributions are twofold. First, we present the designed circuit
relevant to humans heart rate. Second, we examine heartbeat
values of our circuit implementation compared to that of a
standard instrument.
Keywordshealthcare monitoring, heartbeat sensor, circuit
analysis, Internet of Things, smart city
I. I NTRODUCTION
Health monitoring is one of smart services in smart cities
that integrate multiple information and communication technology interacting with people to enhance quality of life.
Health monitoring application is not only for structural health
monitoring [1] but also for biomedical engineering [2]. Health
monitoring can be a key performance indicator reflecting a
cooperative between a smart city and quality of life. The
possibility of two events in a smart city can be as follows.
In structural health monitoring, when vibration of a bridge
surpasses a threshold, alarm messages will be reported to
emergency department via wireless sensor networks [3], [4]
and [5]. In healthcare monitoring, while people are exercising,
their heartbeats are sent to a healthcare centre via connected
devices. Based on two such events of health monitoring systems, sensor circuit and signal analysis are important because
they can improve accuracy of measurement. In order to narrow
down the list of possible events, let this paper focus on
heartbeat measurement.
Heartbeat measurement is a contemporary topic of health
monitoring research since it directly affects quality of life.
Heartbeat can be measured by counting the number of contractions of the heart per minute. This depends on the bodys
physical needs, including the need to absorb oxygen and excrete carbon dioxide. It is normally close to the pulse measured
at any peripheral point. Heartbeat is obviously relevant to
everyday activities such as physical exercise, sleep, anxiety,
stress, illness, and ingestion of drugs. This motivates many
researchers to study how it works. Even though existing work
proposed a variety of heartbeat measurement techniques but
they incompletely explain in terms of analog and discrete
signal analysis.
This paper aims to study continue analog and discrete

signals and provides explanation of how a heartbeat sensor
circuit works. A design of heartbeat measurement consists of
a heartbeat sensor, a noise filter, a signal amplifier and digital
signal processing. There are two steps for circuit and signal
analysis. In a first step, when a finger is placed on the heartbeat
sensor, heartbeats are detected and their signals fed into the
noise filter and the required signal amplifier, respectively. At
this point, heartbeat signals are considered as non-discrete
signals or analog signals since they have a theoretically infinite
resolution. In a second step, heartbeat signals are transformed
from non-discrete signals to discrete signals and then they are
manipulated in order to reduce noises and finally to obtain
heart rates in a unit of beats per minute (BPM).
The rest of this paper is organized as follows. Section II
describes related works of heartbeat measurement research,
Section III presents circuit and signal analysis. Section IV
evaluates our implemented circuit and shows experiment results. Section V concludes the paper.
II. R ELATED WORKS
A trend of Internet of Things motivates a number of
researchers to develop hardware, software, and applications of
healthcare monitoring systems. Researchers [6], [7] presented
design and implementation of the remote healthcare monitoring system. This innovation assists patients to report their
heartbeats from residences to a hospital. Authors [8] reviewed
heartbeat measurement by using smart phones. However, they
were incomplete in explanation of heartbeat sensor circuit
design. For example, a designed pulse sensor circuit in [7]
lacks of a noise filter, this easily leads to interference and
wrong data. In [9] and [10], authors proposed a heartbeat
sensor prototype. For [9], a proposed active low-pass filter
is an opamp having feedback of a resistor of 10 k and a
capacitor of 100 nF. Referring to equation (2), both resistor
and capacitor values give the cutoff frequency close to 16,000
Hertz that is unsuitable for reducing heartbeat noises. For
[10], their explanation gave a wrong formula of gain in
each state. Work [2] provided a comprehensive framework
for heartbeat measurement during intensive physical exercise
but their framework was operated in digital signal processing
under an assumption of noise free environment in a frequency
band from 0.4 to 5 Hertz.
Fig. 1.
Fig. 2. Input and output of heartbeat signals reshaped by (a) the 1st state
of the HPF, (b) the 1st state of the BPF, (c) the 2nd state of the HPF and (d)
the 2nd state of the BPF (heartbeat pulses)
Heartbeat sensor circuit
III. C IRCUIT AND SIGNAL ANALYSIS

This section presents the circuit and signal analysis. The
presented circuit can be divided into analog and digital signal
processing. To analyze and measure heartbeat signals, explanation can be classified into non-discrete and discrete signals
as follows.
A selected low-pass filter is an active opamp filter circuit

that passes signals with a frequency lower than a certain cutoff
frequency and attenuates signals with frequencies higher than
the cutoff frequency. The aim of the LPF is to block any
possible noises with higher than a certain cutoff frequency
such as thermal noises, interference from daylight lamp and
electromagnetic interference from power source.
A. Non-discrete signal analysis

In detecting heartbeat signals, a photo diode is applied for a
light detector while an Infrared Light Emitting diode (LED) is
applied for a light source. An infrared LED is a good choice to
detect heartbeat because the absorption of an infrared light by
oxygenated hemoglobin (HbO2) and reduced hemoglobin (Hb)
is much higher than that by other human tissue or water [11].
A heartbeat sensor consists of a photo diode and an Infrared
LED. When a finger is placed on the heartbeat sensor, a photo
diode absorbs infrared light, produces current flow and results
in blur heartbeat pulses. This blur heartbeat pulses are not
only tiny in a scale of millivolts but also consist of multiple
frequencies including noises. In order to isolate pulses, noise
filter and amplifier circuits are required to reduce noises and
increase the magnitude of heartbeat signals.
To get rid of noises, a band pass filter (BPF) consisting
of a high pass filter (HPF) and a low pass filter (LPF) is
implemented. Since heart rates generally are between 60 to
180 BPM, therefore the corresponding frequency is between
180
60
60 = 1 to 60 = 3 Hertz. A selected high-pass filter is a
passive RC filter circuit that passes signals with a frequency
higher than a certain cutoff frequency and attenuates signals
with frequencies lower than the cutoff frequency. The aim of
the HPF is to block direct current (DC) signals and DC ripples.
fc,HP F 1 =
1
2R3 C1
(1)
fc,LP F 1 =
1
2R4 C2
(2)
To increase the magnitude of heartbeat signals, an op amp

is not only configured for the active low-pass filter but also
for its non-inverting amplifier function by giving the output
feedback directly to the inverting input and applying the input
signal to the non-inverting input.
Av = 1 +
R4
R5
(3)
Combination of equation (1) and (2) is a frequency range

of the BPF. A voltage gain of an amplifier circuit can be
expressed in equation (3). Fig. 1 from point A to C, where
R3 = 68 k, R4 = 680 k, R5 = 6.8 k and C1 =
1 F, C2 = 100 nF, presents a first state of the BPF that
provides a cut off frequency of 2.34 Hertz for both HPF and
LPF and produces a voltage gain of 101. Similarly, Fig. 1 from
point C to E where R6 = 68 k, R7 = 680 k, R8 = 6.8 k
and C3 = 1 F, C4 = 100 nF, this is a second state of the BPF
that provides the same results of the first state. Analytically
from point A to E, the voltage gain is 101101 = 10,201 and
the BPF provides a cut off frequency of 2.34 Hertz for both
HPF and LPF.
Fig. 2 shows measurements at point A, B, C, D, and E
of Fig. 1. Heartbeat signals at point A, B, C, D and E are
gradually changed from vague to sharp pulses as illustrated
Fig. 3.
Discrete-time signals
as Fig. 2(a), (b), (c), and (d), respectively. Signals in millivolt

levels at point A can be maximized to 3.25 volts at point
E. In Fig. 2(d), a heartbeat duration is approximately 700
milliseconds or the inversion of its pulse period is 1.43 Hertz.
As seen a value of 3.25 volts and that of 1.43 Hertz, these
measured values correspond to our circuit analysis where Av
is 10,201 and the frequency band is 1-3 Hertz.
B. Discrete signal analysis
After getting signals at point E illustrated in Fig. 2(d),
heartbeat pulses run into digital signal processing that consists
of sampling data, reducing noises, detecting peaks, finding
Interbeat interval (IBI) and computing a heart rate in BPM.
In sampling data, heartbeat pulses are conversed from nondiscrete to discrete-time signals by using an ATMmega328micro controller. A sampling rate can be determined by its
timer. Discrete-time signal depends on resolution of Analog
to Digital Conversion (ADC). Since ATMmega328 provides a
10 bit ADC, therefore a discrete-time signal can be quantized
5V
into 0 to 1023 levels and its step size equals to 1024
4.88
millivolts. An example of 21 discrete-time signals is illustrated
in Fig. 3. Discrete-time signals generally can be shown in
equation (4).
y=
n
X
yi
(4)
i=1
where yi is a discrete-time signal and n is the total number

of data sampling.
In reducing noises, a discrete-time signal yi is compared
to a certain threshold c in each heartbeat pulse. If yi is less
than c then yi is considered as a discrete noise otherwise yi
is defined as xj .
x=
m
X
xj
(5)
j=1
where xj is a peak candidate and m is the total number of

peak candidates in a heartbeat pulse. In Fig. 3, for an example,
peak candidates are a group of y5 , y6 , y7 and a group of y16 ,
y17 , y18 in the first and second heartbeat pulse, respectively.
In detecting peaks, consecutive peak candidates are compared to each other. When xj+1 is less than xj , this xj is
considered as the highest discrete-time signal or simply called
a discrete-peak. After detecting M heartbeat pulses, a time series of discrete-peaks can be written as p = [p1 , p2 , . . . , pM ]T .
An IBI is the difference between two consecutive peak

times. Once a discrete-time peak p is detected, a corresponding
peak time h is recorded. A periodic time series of peak times
can be written as h = [h(1), h(2), . . . , h(M )]T and its firstorder difference can be defined as h0 = [h(2) h(1), h(3)
h(2), . . . , h(M ) h(M 1)]T . Alternatively
2 3
h1
6 h2 7
6
7
h0 = 6 . 7
(6)
.
4 . 5
hM
where h1 is [h(2) h(1)], h2 is [h(3) h(2)], and hM is
[h(M ) h(M 1)], respectively.
In computing a heart rate, the micro controller counts the
number of IBIs within amount of time and then estimates
the number of IBIs in one minute. A time series of IBIs
can be written as h0 = [h1 , h2 , . . . , hM ]T and mapped into
a trajectory L K matrix [2].
2
3
h1
h2
...
hK
6 h2
h3
...
hK+1 7
6
7
H =6 .
(7)
.
.. 7
..
..
4 ..
.
. 5
hL
hL+1
...
hM
A trajectory matrix H = [H1 ,H2 ,. . . ,HK ] can be decomposed into the multi-dimensional series H1 , H2 , . . . , HM with
vector Hi = (hi , . . . , hi+L 1 ) where K = M
L + 1.
Alternatively, H can be written in (8).
H=
d
X
Hi ,
d = min {L, K}
i=1
(8)
where i is an integer number between 1 to d.

There are two steps in estimation of the number of IBIs in
one minute. That are finding an average of IBIs and doing the
rule of three in arithmetic for computing a heart rate. To find
average of IBIs, the vector Hi are assigned into g groups. That
is, the set of indices {1,. . . , d} is partitioned into g disjoint
subsets {I1 ,...,Ig }, g d and
H=
g
X
HIp
(9)
p=1
with a vector HIp = (h1 , . . . , ht ) where t 2 g. All member

values in the vector HIp are calculated for an IBI average, i .
t
1X
hi
t i=1
(10)
In finding a heart rate, if a unit of an IBI average, i ,

is milliseconds then a heart rate, HR, can be computed by
equation (11).
HR =
60, 000
i
BPM
(11)
TABLE I
P ERFORMANCE OF HEARTBEAT MEASUREMENT
Index
i
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
Fig. 4. Comparison of our implemented heartbeat circuit and a standard

heartbeat instrument in a first round
IV. P ERFORMANCE MEASUREMENT

A. Implementation
Based on our circuit and signal analysis, we implemented
our idea into an analog heartbeat circuit and an Arduino
MAGA board. In digital signal processing of heartbeat measurement, a sampling rate is two milliseconds per a discrete
signal. In computing IBIs based on the LK matrix condition
where L M/2 and K = M L + 1, we set L, K and M
are 25, 10, and 50, respectively. For simplicity, let d, g, t be
the same value.
IBI
i
(ms)
750
738
768
732
740
722
714
724
724
760
706
722
722
750
732
740
700
700
750
750
732
698
706
714
698
Measured
HRo
(BPM)
80
81
78
82
81
83
84
83
83
79
85
83
83
80
82
81
78
78
80
80
82
86
85
84
86
Standard
HRs
(BPM)
80
80
81
82
82
81
82
84
83
82
82
82
81
82
83
80
79
79
80
82
83
85
86
85
85
Absolute error
|HRs HRo |
(BPM)
0
1
3
0
1
2
2
1
0
3
3
1
2
2
1
1
1
1
0
2
1
1
1
1
1
B. Experiment results
During experiment, when a finger is placed on the heartbeat
sensor, a first valid heartbeat can be detected after 5-10 seconds. Then their readings are periodically changed according
to human heart rates. In Fig. 4, heart rates of our implemented
circuit HRo are compared to that of a standard instrument
HRs [12]. A round of heartbeat measurement consists of W =
25 readings. In a first round, 25 readings are shown in Table I
where a mean of absolute errors is 1.28 BPM and standard
deviation, , is 0.89. IBIs are between 698 to 768 milliseconds
(1.43 - 1.30 Hertz). A time series of absolute errors H can be
written as equation (12).
E=|
W
1 X
HRs
W i=1
W
1 X
HRo |
W i=1
Fig. 5.
Error means and standard deviations
of heartbeat measurement
(12)
This experiment results are corresponding to our analog

circuit analysis referred to Fig. 2(d), where a heartbeat duration
is approximately 700 milliseconds. The summery results of
Table I is shown in a first left of an error bar graph in Fig.5.
In our experiments, we repeatedly measured heartbeats of 10
rounds as illustrated results in Fig. 5. The maximum and
minimum of absolute errors are 2.67 BPM and 0.15 BPM,
respectively. An average of error means is 1.14 BPM. These
experiment results imply that a maximum error is less than 3
BPM and average of error means is less than 2 BPM.
V. C ONCLUSION
This paper analyzed non-discrete and discrete signals processing in heartbeat measurement. The analyzed circuit consists of a heartbeat sensor, a noise filter, a signal amplifier
and digital signal processing. Our contributions are twofold.

First, we present the designed circuit relevant to humans
heart rate. Second, we examine heartbeat values of our circuit
implementation compared to that of a standard instrument. In
experiment results, a maximum error is less than 3 BPM and
average of error means is less than 2 BPM.
ACKNOWLEDGEMENT
This research was supported by Faculty of Engineering, Burapha University, Thailand under the contract number 15/2555.
The authors gratefully acknowledge the contributions of all
committee members for their valuable suggestions.
R EFERENCES
[1] Z. Zou, Y. Bao, F. Deng, and H. Li, An approach of reliable data
transmission with random redundancy for wireless sensors in structural
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
health monitoring, IEEE Sensors Journal, vol. 15, no. 2, pp. 809818,
Feb 2015.
Z. Zhang, Z. Pi, and B. Liu, Troika: A general framework for
heart rate monitoring using wrist-type photoplethysmographic signals
during intensive physical exercise, IEEE Transactions on Biomedical
Engineering, vol. 62, no. 2, pp. 522531, Feb 2015.
A. Boonsongsrikul, S. Kocijancic, and S. Suppharangsan, Effective energy consumption on wireless sensor networks: Survey and challenges,
in Information Communication Technology Electronics Microelectronics
(MIPRO), 2013 36th International Convention on, May 2013, pp. 469
473.
A. Boonsongsrikul, S. K. Park, and S. H. Shin, An adaptively balancing
workload protocol on query trees for maximizing lifetime in sensor
networks, in Advanced Communication Technology (ICACT), 2012 14th
International Conference on, Feb 2012, pp. 195200.
A. Boonsongsrikul, K. suk Lhee, and M. Hong, Securing data aggregation against false data injection in wireless sensor networks, in
Advanced Communication Technology (ICACT), 2010 The 12th International Conference on, vol. 1, Feb 2010, pp. 2934.
H. Mansor, S. S. Meskam, N. S. Zamery, N. Q. A. M. Rusli, and
R. Akmeliawati, Portable heart rate measurement for remote health
monitoring system, in Control Conference (ASCC), 2015 10th Asian,
May 2015, pp. 15.
S. R. Mohana and H. V. R. Aradhya, Remote monitoring of heart
rate and music to tune the heart rate, in Communication Technologies
(GCCT), 2015 Global Conference on, April 2015, pp. 678681.
M. T. I. Papon, I. Ahmad, N. Saquib, and A. Rahman, Non-invasive
heart rate measuring smartphone applications using on-board cameras:
A short survey, in Networking Systems and Security (NSysS), 2015
International Conference on, Jan 2015, pp. 16.
M. Ghamari, C. Aguilar, C. Soltanpur, and H. Nazeran, Rapid prototyping of a smart device-based wireless reflectance photoplethysmograph,
in 2016 32nd Southern Biomedical Engineering Conference (SBEC),
March 2016, pp. 175176.
M. A. Miah, M. H. Kabir, M. S. R. Tanveer, and M. A. H. Akhand,
Continuous heart rate and body temperature monitoring system using
arduino uno and android device, in Electrical Information and Communication Technology (EICT), 2015 2nd International Conference on,
Dec 2015, pp. 183188.
G. Pang and C. Ma, A neo-reflective wrist pulse oximeter, IEEE
Access, vol. 2, pp. 15621567, 2014.
S. Smiths Medical PM. (2008, May) Bci3301hand-held pulse oximeter.
[Online]. Available: https://goo.gl/25yKdn
Design of Zero Energy Consumption Home

Using Stand Alone Solar System
Nidchabendha Chandanachulaka and Werachet Khan-ngern
AbstractThis paper presents the design of zero energy

consumption home (ZECH) on a stand-alone photovoltaic
system. System structure composes of three main parts: 2x300 W
PV system, 2880 Wh energy storage of battery and charging
system, and the 1,000 W inverter for AC use. The load
requirement which is less than 1,500 Wh per day located in
Bangkok.3 scenarios of load requirement on full sunshine, half
day sunshine and no sunshine conditionsare proposed including
themodel on the appropriate proportion of generated energy :
load energy : storage energywhich becomes 2:1:1. Collected data
is monitored via a Web server application in real-time. The
analysis of the zero energy consumption has been done all the
solar conditions.
Keywordszero energy consumption home (ZECH)
Photovoltaic; stand-alone system; utilization pricing Ratio
I. INTRODUCTION
The term zero energy buildings, recognized as net zero
energy buildings are about the buildings thathave zero carbon
emissions on annual basis. The daily energy requirements and
generate energymust be minimized by using renewable
resources [1].
The concept of zero energy buildings is completely about
minimizing the required energy use, particularly energy that
generated from fossil fuel and implementing the use of
renewable energy resource to meet the energy demand.
supported by Danny H.W. Li et al [2].
Energy consumption can be minimized by applying new
technologies in the buildings such as wall materials, insulation
materials, high efficient air conditioning system or solar
lighting which lead to meet zero energy building designs [3].
Farajallah Alrashed [4] mentioned that zero energy buildings
actually rely on renewable energy to meet their energy needs,
therefore, it isimportant to have sufficient and appropriate
renewable resources to sustain the energy demands and
photovoltaic or PV seems to be one of the most sustainable
energy generation technologies for zero energy building [5].
which do not consume any fossil fuels and renewable energy

must be used in zero energy consumption home in order to
replace the use of fossil. In addition, the appropriate size of
energy storage is required to preserve the energy that
generated from renewable resources. However, there are many
studies have presented that an off-grid PV system is viable in
rural areas and remote locations.
Guda H.A. et al [7] provided a case study of an off-grid
PV system for the 3 bedroom bungalow in Bauchi, Nigeria,
and the residential daily energy demand was 6.5 kWh per day
and the success of each off-grid PV system depended on
different factors such as weather condition, solar irradiance,
load profile, and geographical location. However, the most
important requirements for a storage for off-grid PV system
are low cost, long life time with low maintenance, high energy
efficiency, self-discharging and easy operation [8].
III. PROPOSED ZERO ENERGY CONSUMPTION HOME
The main components of zero energy consumption home
system consists of PV arrays, solar charger controller, battery,
and inverter as presented in Figure 1.
A. PV arrays
Photovoltaic power generation is the process that directly
generates the electricity from the daylight, therefore, PV
arrays is the most important component in the system. 300 W
Polycrystalline Silicon has been selected for this designed
system.
Solar Cell
Controller
Battery
Inverter
Fig. 1. The zero energy consumptionhome system components.
II. ZERO ENERGY CONSUMPTION HOME CONCEPT

Since M. Iqbal expressed that zero energy buildings may
or may not be grid connect[6], it is possible to state that zero
energy consumption home could be an off-grid buildings
Energy
Usage
Cabinet
B. Solar charger controller
The solar charger controller manages the power flow from

the PV arrays to battery and to prevent the battery from being
overcharged [7].
C. Battery
The battery is charged during the sunshine hours and used
to supply the load during the night time and no daylight hours.
It is necessary to have the appropriate size of energy storage
that used in the of-grid or stand-alone side.
D. Power inverter
The main activity of the power inverter is to convert the
direct current (DC) into the alternating current (AC) at 220
volt. 1000 W Pure sine wave power inverter is selected to use
because it suites for the continuous load and could generate
AC load directly.
E.Energy demand for a ZECH
8
7
6
5
4
In addition, this zero energy consumption home system

also concerns the utilization-pricing ratio. The result from this
paper will be used in the small residential house that require
the electricity for basic needs and this system must be worth
for investment as this paper targets that the break-even must
be reached within 15 year.There are 3 scenarios, depending on
the situations of sunlight, which are 1) full-day sunlight, 2)
half-day sunlight and 3) no sunlight.
IV. THE ZERO ENERGY CONSUMPTION HOME
SYSTEM SETEUP AND RESULTS
(a) Pilot project home.
(b) system monitoring.
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Oct
Nov
Dec
AVERAGE SOLAR
RADIATION (KW/M2-DAY)
During 2010-2014, Bangkok has the average solar

radiance of 5.84 kW/m2-day as shown in Figure 2 [9]. It can
be observed that the daily solar radiation in Bangkok has a
peak in April 2012, while the daily temperature reached 40
degree Celsius. Also, the daily solar radiation has slightly
dropped during the rainy reason, July October. .
house with less than 1,500 Wh energy demand and without

any high energy consumption usages such as air condition or
washing machine. The proposed residential house isan offgrid PV system for electrification, located in Meanburi
district, Bangkok. Its size is 3m x 7.5 m with AC supply
system. In addition, this paper targets that the break-even
must be reached within 15 year.The load requirement for this
residential house is 1.262 kWh/day as indicated in Table 3.
MONTHS
Fig. 2. The average solar radiation in Bangkok during 2010-2014.
Equipment
No. of
equipment
Power of
Equipment (W)
Daily
use (h)
Light 1
Light 2
Light 3
Fan
1
1
1
1
12
12
12
58
1
1
12
18
Computer
50
Total
Energy
required
(kWh/d)
0.012
0.012
0.144
1.044
(c) data analysis system.
0.05
1.262
TABLE I. LOAD REQUIREMENTS
Electric energy for their basic needs such as lights, TV for

news, such as a fan during hot weather are needed. They use
electricity just a few hours at night. Not only people who are
living in rural areas, low income people in Bangkok also
concern about how to minimize the use of energy in their
household. Therefore, this paper focuses ona small residential
(d) hardware of data analysis system.
(e) example of data monitoring.
Fig.3.Zero energy consumption home system.
The data analysis system of zero energy consumption

home composes of sensors, CPU, communication and
monitoring are shown in Figure 3. The zero energy
consumption home size of 22.5 m2 shown in Figure 3a where
the PV is attached at the south-north direction with a slope of
14 degree. The energy and temperature parameter are
monitoring in Figure 3b. The diagram and prototype of data
analysis system are shown in Figure 3c and 3d respectively
where the PV parameters of DC voltage, DC current, PV
temperature, AC voltage and AC current from inverter will be
calculated in CPU using DSP33F. The monitoring of key
information of energy is shown in Figure 3e. The dada can be
monitoring at any location via the internet system. The remote
data can be nearly real-time analyzed.The zero energy
consumption model are expressed as follows:
Having calculated Energygen on the sunshine day, there

are 2 PV arrays, each array can generate 300 W. It is observed
that PV array can generate the electricity for 4.23 hours/day in
Bangkok[18]. Basically, PV arrays could generate 2,538 Wh.
After the electricity passes the controller, with 95% efficiency,
the remaining electricity becomes 2,411Wh. Next,
Energystorageis calculated by deduct required load, which is
1,086 Wh as shown in Figure 4a and the forecasted load
profile is shown in Figure 4b.
Solar
Cell
2,538
Wh
Controll
er
DC-DC
2,411
Wh
Inverter
DC-AC
1,325
Wh
Load
1,262
Wh
1,086
Wh
Battery
2,880
Wh
Loss
127
Wh
(a)
Loss
66 Wh
Calculation.
A. Equations
As the concept of off-grid PV system, the electricity that
generated from PV power generation must be sufficient to
support the energy demand and must be able to preserve in the
battery in order to use the electricity when the PV system
cannot generate the electricity because of no sunshine. The
equation for the generated energyis shown in (1)
(b)
Energygen = Energyload + Energystorage
(1)
Fig. 4. Full-day sunlight.
Energyload has been calculated from the load profile as

shown in (2)
x (P1xt1 + P2xt2 + Pnxtn) = (Pt)load (2)
Energygen has been calculated from peak power multiplied
by voltage and time that PV arrays could generate the
electricity from the sunshine hour as shown in (3)
PVpeak x tavg = (Pt)gen
(3)
This paper assumes that PV arrays can generate the

electricity for 4.23 hours/day[17]. Energystoragehas been
calculated from Energygenminused byEnergyload as shown in
(4)
(Pt)gen- (Pt)load = (Pt)storage
Forecasted load profile.
C. Scenario 2: limited or half-day sunshine

In this scenario, PV arrays are supposed to generate
electricity for 2.115 hours day, so PV could generate 1,272
Wh. After electricity passes the controller, the remaining
electricity becomes 1,208 Wh which is not enough for the
1,262 Wh load requirement. It is necessary to use 117 Wh
from the save energy in battery. If battery is full charged and
sunlight is available only few hours, it can use save energy in
the battery to support load requirement for 19.4 days as shown
in Figure 5a and 5b.
Solar
Cell
1,272
Wh
(4)
Noted that, losses in the system such as Ohmic loss has not
been calculated yet.
B. Scenario 1: full-day sunshine
Controlle
r
DC-DC
1,208
Wh
Loss
64 Wh
117
Wh
Battery
2,880
Wh
(a)
Calculation.
Inverter
DC-AC
1,325
Wh
Loss
66 Wh
Load
1,262
Wh
(b)
Wh per day in Bangkok, based on the concept of zero energy

consumption home (ZECH) was presented. This design
system shows that the possibility of the stand-alone PV system
for a small residential home can be done successfully in all
weather conditions. The model of appropriate proportion of
generated energy: load energy: storage energyof 2:1:1 ratio is
proposed. In case of a full day sunshine is operated, battery is
run on the storage mode. In a half day sunshine, the battery
can be run either storage mode or discharge mode depended
on the load condition. The real situation of no sunshine occurs
for two days, the battery discharges up to 1,800 Wh. This
storage energy system can be sustainability operated. It is
confirmed that three scenario of sunshine, the proposed model
can be exactly run in the real situation
Fig. 5. Half-day sunlight.
D. Scenario 3: no sunshine
In the raining day that PV array cannot generate electricity,
it is necessary to use the energy that saved in battery. If the
battery is full charged, it is possible to use the energy for 1.7
days as shown in Figure 6aand the forecasted load profile is
shown in Figure 6b.
Inverter
DC-AC
1,325
Wh
Controll
er
DC-DC
0 Wh
Solar
Cell
0 Wh
Load
1,262
Wh
ACKNOWLDGMENT
Authors would like to thank Dr. Bhirasak Vorasutharosoth
to permit and support the installation of PV system, battery,
inverter and data analysis system to collect useful data in his
meditation home.
REFERENCES
[1]
1,325
Wh
Battery
2,880
Wh
Loss
0 Wh
(a)
[2]
Loss
66 Wh
Calculation.
[3]
[4]
[5]
(b)
(c)
Fig. 6. No sunlight.
[6]
E. Energy ratio
Full-day sunlight is the best case that used to find the
energy ratio. This system can generate electricity 2,538 Wh
with 1,262 Wh load requirement and the remaining 1,086 Wh
is saved in the battery. Without concerning other losses in the
system, such as Ohmic loss, Energygen : Energyload :
Energystoragewill become 2:1:1.
V.
CONCLUSION
[7]
[8]
[9]
Angeliki Kylili, Paris A. Fokaides, "European smart cities: The

role of zero energy buildings," Sustainable Cities and Society,
vol. 15, pp. 86-95, 2015.
Danny H.W. Li, Liu Yang, Joseph C. Lam, "Zero energy
buildings and sustainable development implications - A
review," Energy, vol. 54, pp. 1-10, 2013.
Tanakorn Wongwuttanasatian, Denpong Soodphakdee,
Narinporn Malasri, Kittichai Triratanasirichai, "A demonstrated
net zero energy building in Thailand: The way for sustainable
development in buildings," Advanced Materials Research, vol.
1119, pp. 741-747, 2015.
Farajallah Alrashed, Muhammad Asif, "Analysis of critical
climate related factors for the application of zero-energy homes
in Saudi Arabia," Renewable and Sustainable Energy Reviews,
vol. 41, pp. 1395-1403, 2015.
A. Scognamiglio, G. Adinolfi, G. Graditi, E. Sarett,
"Photovoltaics in net zero energy buildings and clusters:
enabling the smart city operation," Energy Procedia, vol. 61,
pp. 1171-1174, 2014.
M. Iqbal, "A feasibility study of a zero energy home in
Newfoundland," Renewable Energy, vol. 29, pp. 277-289,
2004.
Guda H.A., Aliyu U.O., "Design of a stand-alone photovo;taic
system for a residence in Bauchi," International Journal of
Engineering and Technology, vol. 5, pp. 34-44, 2015.
R.K. Akikur, R.Saidur, H.W. Ping, K.R. Ullah, "Comparative
study of stand-alone and hybrid solar energy systems suitable
for off-grid rural electrification: A review," Renewable and
Sustainable Energy Review, vol. 27, pp. 738-752, 2013.
DEDE, 2015.
The stand-alone photovoltaic (PV) system for a small

residential home, which load requirement is less than 1,500
Design of Automated Guided Vehicle for Industry

Krit Duangprom, Jakkarin Promjan, Chanathip Akkaakara, Chananan Harisombatand Werachet Khan-ngern
AbstractThis paper presents a design a prototype of

automated guided vehicle (AGV) for industry. This automated
guided vehicle is designed to meet the requirement of automatic
moving in the industry or working area with safety when any
person or object obstruct on the vehicle path with ultrasonic
sensor. The aluminum guided tape is selected due to the
simplicity of use and low cost. Six sensors can detect the metal
tape using inductive proximity sensors. The speed is limited at 5
km/h to comply with safety requirement of safety standard in the
industry. This AGV can be driven by two brushless direct
current motors (BLDC) 750 W, 700 rpm, 36 Vdc and 50 poles. 36
V 10 Ah lithium iron phosphate battery is the energy source for
this vehicle can drive up to 8 hours, as it has light weight and
high efficiency.The vehicle payload can be used up to 80 kg.
ATMega2560, ATTiny85, nano328 and dSPIC30F2010 are
control devices for this vehicle with software to provide the full
AGV operation. The AGV prototype is tested with a good
response and comment from the industry.
KeywordsAutomated guided vehicle; Brushless direct current
motor (BLDC motor); Inductive proximity sensor; Guided tape;
Battery monitoring
I. INTRODUCTION
At present, automation system is popular in the industry
due to the effect of competition in the market. An important
factor is to deliver the products in time with no man day labor
cost. AGV safely transport is operated without human
intervention within production, logistic, warehouse, the clear
way to reduce costs and to increase efficiency and
profitability. If considering the cost of process production is
dominated by materials handling. Therefore, we can use the
AGV to work in the industry. It can increase production
capacity on process, reduce labor cost and also provide
flexibility in working.
The AGV is designed as a prototype using brushless DC
motor which has high efficiency, smaller than induction
motor, light weight and less maintenance. In this paper the
objectives are to improve precision of navigation system, to
improve AGV control, monitoring system and to improve
reliability of AGV. For the scope of work using gear wheel
hub brushless DC motor 750W rated power, 100-700 rpm 36
Vdc. The AGV is functioned and control by microcontroller
ATMega2560, ATTiny85, Nano328 and dsPIC30F2010 using
inductive proximity sensor to detect path. The battery

consumption is measured and show their energy status.
II. PROPOSED AGV SYSTEM DESIGN
The AGV system design is starting from the function and
specification of industry need. Safety and good performance
are basic requirement of AGV and most of vehicles used in
industry. The key function of this prototype are as follows:
1. pay load up to 80 kg.
2. track on guided line which start-stop at each station by
specified on program.
3. can be operated as automatic or semi-automatic
4. stop if there is any obstruction on the guided line
5. warning and monitoring of AGV operation should be
prepared
Three main parts of AGV system are composed of input
devices, CPU and output devices. The input parts are
inductive sensors, ultrasonic sensors, joy stick and guided path
to send data the CPU. The CPU processes all data to calculate
and generate the proper signal and information to the output
parts such as motor control and monitoring system as shown
in figure 1.
Figure 1. Three min part of AGV
A. Central Processing Unit

AGV uses microcontroller ATMega2560 for receiving
the signal from inductive proximity sensor and then sending
the signals to control two BLDC motors through motor drive

control. While ultrasonic sensor detects obstacle object, the
signal will be sent toATTiny85 to generate the stop signal for
emergency stopping the AGV.The main controller is managed
via dSPIC30F2010.The CPU board is shown in figure 2 a.
III. RESULT OF THE AGV PERFORMANCE TEST

Theexperiment test is done based on the safety and
performance as shows in table 1.
Safety have bene tested such as push button for emergency
switch. The 30 cm checking for any object obstruction is
working properly. The operating status can be shown both
lighting and sound. Battery and charging are monitoring and
simplified.
a. CPU
b. BLDC motor drive
Figure 2 CPU and drive of AGV
B. BLDC Motor Drive

Two bruchless DC mortor (BLDC) are controlled via
dsPIC30F2010. Control signals drives main MOSFETson
unipolar PWM inverter at switching frequency at 20kHz via
IR-2101 [1-2].2 kW inverter shown in figure 2 b.BLDC motor
can run at 700 rpm, 36 Vdc and 50 poles. 36 V 10 Ah lithium
iron phosphate battery is the energy source for the AGV.
C. Sensors
Two types of sensors are designed. First sensors are
guided sensor using inductive proximity sensors for noncontact detection of metallic objects. Their operating principle
is based on a coil and oscillator that creates an electromagnetic
field in the close surroundings of the sensing surface. The
presence of a metallic object (actuator) in the operating area
causes a dampening of the oscillation amplitude. The rise or
fall of such oscillation is identified by a threshold circuit that
changes the output of the sensor. The 4 mm. operating
distance of the sensor is confirmed the track detection. The
other sensors are ultrasonic to sense an object to stop avoid
any damage in AGV crash. Figure 3 a. shows the sensors
installation. Figure 3 b. shows AGV can achieve the travel
pass the 45 degree corner and small curve.
The good AGV performance are passed all track corner

such as 45, 90 corner and 10 cm round curve. The use of
glued aluminum tape provide a good detection at 4 mm. gap.
The AGV speed can be controlled and limit up to 5 km/h due
to the industry regulation. 1 horse power of two driving BLDC
motors are enough for 80 kg payload. User can program the
start/stop at each station for instance station x requires 20
second stop for the process while station y require 10 second
to top. The setup can be done as the factory requirement.
Automation home return also can be done.
The AGV is demonstrated at one company in
Prathumthani. Thailand. The feedback is positive and some
comments on material selection is given.
safety
performance
emergency stop
45, 90 corner
30 cm ultrasonic stop
10 cm round curve
siren mode
5 km/h speed with load

adjustment
warning sound
80 kg payload
battery status
programmable station stop
simplicity of charging
home/end setup
IV. CONCLUSION
The AGV are designed to meet the safety and performance
of industry requirement. Aluminum tape is a solution of
tracking guide path with low cost. The AGV system designed
is proposed. The prototype need to be improved on the
material as comment from the industry feedback
REFERENCES
a. AGV sensors
Figure 3 BLDC motor drive
b. 45 degree and small curve
[1]
Ronnakit Jitjamsai Drive system for car mode ofamphibious electric

vehicle, Engineering Thesis, ElectricalEngineering, King Mongkuts
Institute of Technology Ladkrabang, 2012.
[2]
Krittipass Piwsantia, Design of brushless DC hubmotor prototype,

Engineering Thesis , Electrical Engineering, King Mongkuts Institute
of Technology Ladkrabang, 2011.
The Implementation of a Wireless-Remote ArduinoCompatible Programmable Logic Controllerwith Smart

User Interface towards Industrial Internet of things
Tossaporn Wetsiri and Wimol San-Um
Intelligent Electronic System (IES) Research Laboratory
Master of Engineering Technology Program, Faculty of Engineering, Thai-Nichi Institute of Technology (TNI)
1771/1, Pattanakarn Rd., Suan Luang, Bangkok, 10250, Thailand.Tel: (+66) 0 2763 2600, Fax: (+66) 0 2763 2900.
E-mail: wetsiri@gmail.com and wimol@tni.ac.th

ABSTRACT
The Industrial Internet of Things (IIoT) typically enables remote
communications via wireless IP devices and interactions between
human and manufacturing processes through sensors and
actuators, resulting in the enhancement of performances and
efficiency as well as the reduction in manufacturing costs. This
paper presents a complete IIoT platform, involving all physical,
communication and application layers. The universal
Programmable Logic Controller (PLC) is controlled by an opensoftware Arduino microcontroller via a high-voltage peripheral
board. The communication between PLC and end users is
achieved by Wi-Fi technology and Message Queue Telemetry
Transport (MQTT) protocol that allow data storage and
monitoring in cloud server. The Graphic User Interface (GUI) is
designed for both a web-based display and an Android
application. Demonstrations on the pneumatic control system are
also included. The proposed system offers a low-cost highefficiency implementation of IIoT for Small-and-Medium
Enterprises (SMEs) in ASEAN countires.
CCS Concepts
CCS Hardware Communication hardware, interfaces and
storage Wireless integrated network sensors.
Keywords
Industrial Internet of Things; Wireless Control; Arduino
Microcontroller; MQTT Internet Protocol; Android.
1. INRODUCTION
Digital transformations have been of much interested by industries
and enterprises as an emergence of intelligent devices, ubiquitous
internet connectivity, big data analytics and cloud computing has
potentially driven new business opportunities, reduced waste and
costs, and eliminated loss regarding human-machine (H2M) or
machine-machine (M2M) activities. As a result from the increase
in digital transformations, the Internet of Things (IoT) has
consequently played an important role as a new segment in
existing platforms. Such an IoT can generally be defined as a
system in which the internet, communicating through Internet
Protocol (IP), is connected to the real world via ubiquitous sensors
and actuators, integrating diverse data sets obtained from physical
sensors and Information Technology (IT) infrastructures in order
to enable intelligent data analytics and control processes.
Recently, the IoT has extensively applied to various industries
such as healthcare [1], transportation [2], agriculture and water
supply [3-4], smart home and city [5-6], and manufacturing [7].
The Industrial Internet of Things (IIoT) is an emphasis of IoT

deployment in industrial automation manufacturing in order to
enhance or create system architectures that are more responsive,
affordable, and effective. Technically, IIoT-based systems enable
remote communications via wireless IP devices and interactions
between human and manufacturing input and output processes
through, for example, sensors, actuators, or robotics. It is
generally known that most industries employ a commercially
available Supervisory Control and Data Acquisition (SCADA)
system to manage the manufacturing process and exploit a
Programmable Logic Controller (PLC) to execute machine
operations. Nonetheless, the industrial application based on such
SCADA and PLC systems is relatively expensive in terms of both
purchasing and maintenance services. The operator with highlyspecific skills is also required as a software designer who can
excerpt maximum potentials of the PLC. In addition, perspectives
on industrial automation has moved towards manufacturing
integration through semantic and standardized approachesfor
transparently exchanging data amongSCADA, Manufacturing
Execution Systems (MES), and Enterprise Resource Planning
(ERP) systems. Consequently, the implementation of the
industrial automation manufacturing based on IIoT is ultimately
required as for a solution of flexible controls through wirelesslyremote system and data storage in cloud server, resulting cost
reduction and high level of optimization and efficiency.
Consideration on industrial automation manufacturing based on
IIoT has led to extensive studies in recent years, ranging from
physical to network and application layers. Norbert Gal [8] has
early proposed a remote monitoring and control platform based on
PLC and its communication module with two particular examples,
i.e. a passive greenhouse and a PLC matrix for industrial
applications. Brizzi P. and et al.[9] has described the IoT
exploitation based on a commercially available Ebbits platform
that provides a middleware infrastructure for integrating industrial
devices and emerging wireless technology, and transforming such
devices into web services that enable seamless integration to
business systems such as MES and ERP. Sergey Popov and
Evgeny Chernyy [10] have particularly developed a service
implementing IoT that delegates controlling and monitoring tasks
over heterogeneous devices to the cloud. The manufacturing
facility of accounting and controlling of energy resources system
has also demonstrated. Felipe Corso and et al. [11] has suggested
a wireless control and communication architecture for PLC based
on data collected from SCADA system. Mircea Murar1 and
Stelian Brad [12] have recently proposed the implementation and
testing of the solution for connectivity between smart equipment
and Android-compatible devices using Wi-Fi technology.
Cloud Server
(b) Communication Layer

MQTT
MQTT
Router
ESP8266 Wi-Fi
Module
MQTT
Router
Router
Web-Based
Application on PC
Android-Based
Application on
Smart Phone or
Tablet
Arduino
Compatible Board
(a)
Industrial
Input / Output
via PLC
(a) Machine-to-Machine Interface
(c) Human-to-Machine Interface
Figure 1. Proposed system architecture of a wireless-remote

Arduino-compatible PLC with smart user interface.
High-Speed Counter
Built-In RS232C
Serial Option Port

RS 485
Built-In USB
Modbus RTU
Pulse Outputs
Inverter Inverter
Figure 2. Control Capabilities and Functionality of the PLC.

This paper alternatively presents a complete IIoT system for
Small-and-Medium Enterprises (SMEs), involving all physical,
communication and application layers. The universal PLC is
controlled by an open-software Arduino microcontroller via a 24V peripheral board. The communication between PLC and enduser is achieved by Wi-Fi technology and MQTT protocol that
allow data storage and monitoring in the cloud. The Graphic User
Interface (GUI) involves both a web-based display and an
Android application. Demonstrations on the control of pneumatic
system are also included.
2. PROPOSED SYSTEM ARCHITECTURE

Figure 1 shows the proposed system architecture of a wirelessremote Arduino-compatible PLC with smart user interfaces,
involving (a) Machine-to-Machine (M2M) interface, (b)
communication layer, and (c) Human-to-Machine (H2M)
interface. It is seen in Figure 1 that, on the one hand, the industrial
input and output ports can be connected directly to the PLC,
which is controlled by Arduino controller. Data collected from the
Arduino controller will be sent to a router via the Wi-Fi module.
The router hence communicates to could server through Message
Queue Telemetry Transport (MQTT) protocol. On the other hand,
the control process and state monitoring can be performed through
either web-based display on a personal computer or an Androidbased application on smart phones or tablets. This paradimg hence
shows the simple idea yet providing a practical IoT platform for
some small-and medium enterprieses in ASEAN countries where
industrial 4.0 is becoming a new approch for their countries.
Details for each part will be discussed in the following sections.
(b)
Figure 3. Proposed Arduino compatable board; (a) Circuit
diagram of a Arduino ATMEGA328P, (b) an implemented
board with full periferals for Industrial Input/Ootput ports.
2.1 Machine-to-Machine (M2M) Interface

This part initially focuses on the PLC functionality. A PLC is
typically a digital computer for automation of typically industrial
electromechanical processes such as control of machinery on
factory assembly lines. PLCs have been employed in many
machines, and designed for multiple arrangements of digital and
analog inputs and outputs. Typically, software for controlling
machine operations is stored in non-volatile memory. The
limitation found in PLCs include the controlling program, which
is a licensed ladder programand the expertise is ultimately
required. Addiitonally, the design flexibility is low limited by
such a program, and therefore no gennerally avaliable such as C
or C++ languages can not be used.
Figure 2 shows the specific type of PLC utilised in this work. The
model is SYSMAC CP-series CP1E, which is the E-type basic
CPU units that provide cost performance and easy application
with basic functionality. This model provides programming,
setting, and monitoring with CX-Programmer. As shown in Figure
2, the connection can be accomplished with computers using
commercially available USB cables and built-in RS-232C. There
are also complete high-speed counter functionality and a versatile
PLC Module
Arduino Compatible Board
Power Supply
Figure 4. The ESP8266Wi-Fi modul module that is

particualrly used in this work.
pulse control for transistor Outputs. The specific feature is the
input/output (I/O) can be expanded through connecting expansion
I/O units. Figure 3 presents the proposed Arduino compatable
board, which is particularly designed as for a PLC perfierals with
embedded controller. This borad eases the end users in IIoT
platforms to control and monitor the machine through this simple
board. The advanges are the program is commonly avaliable and
can be written as a simple bolck diagrams through the use of
Arduino libraly. The number of I/O can be expanded by adding
I2C bus.
Wi-Fi
Neumatic Switches
Figure 5. The overall system design, including a PLC, an

Arduino Compatible board, a Wi-Fi module, a power supply,
and a neumatic swutches.
Figure 3 (a) shows the ciruit diagram of the chosen Arduino

ATMEGA328P, which is a high-performance Atmel picoPower 8bit AVR RISC-based microcontroller. Technically, this
microcontroller combines a 32-KB ISP flash memory with readwhile-write capabilities, a 1024-B EEPROM, 2-KB SRAM, 23
general purpose I/O lines, 32 general purpose working registers,
three flexible counters with compare modes, internal and external
interrupts, serial programmable USART, a byte-oriented 2-wire
serial interface, SPI serial port, a 6-channel 10-bit A/D converter,
programmable watchdog timer with internal oscillator, and five
software selectable power saving modes. This device operates
between 1.8-5.5 volts. Figure 3 (b) shows the implemented board.
The poser spuuply is 24V, regulating 3.3V for the WIFI module
and 5V for the microcontroller. The I/O ports connect to the relay
and a regultaed 24V for PLCs. The control can be done by
Arduino program, and thereofre no licenced ladder program is
required. The circuit is fabricated with full periferals, and is ready
to connected to the PLCs without any extra devices.
(a)
(b)
2.2 Communication Layer

The cominucation layer involves a Wi-Fi connection with the
router and the MQTT protocal for clound server connection. In
this work, the Wi-Fi module is a ESP8266 as shown in Figure 4,
which has been designed for mobile, wearable electronics and
Internet of Things applications with the aim of achieving the
lowest power consumption with a combination of several
proprietary techniques. The Wi-Fi module is connected to the
router. The MQTT is utilised aftwerwads to connect to the clond.
As the IoT requires standard protocols, the MQTT is an open
standard and is better suited to constrained environments than
HTTP, providing mechanisms for asynchronous communication,
runnig on IP, and having a wide range of implementations. MQTT
offers flexibility in communication patterns and performs purely
as a pipe for binary data. It has been designed as an extremely
lightweight publish or subscribe messaging transport. This work
stores data in the cloud server and communicate to the end-user
via MQTT as well.
(c)
(d)
Figure 6. The display on the Android Application, (a) a sleep

mode (No machine is ON), (b) all machines is ON, (c) a virual
operating machine of 20 inputs, and (d) the setting functions.
2.3 Human-to-Machine Interface
ACKNOWLEDGEMENTS
The data involving the control signals and the monitroing states
are sent to the router via MQTT protocal. The data is connected to
the samrt devices such as tablets or smart phones through the
integrated Wi-Fi module. Figure 5 shows the overall system
design, including a PLC, an Arduino Compatible board, a Wi-Fi
module, a power supply, and a neumatic swutches. As aresult of
Android designs, Figure 6 shows the display on the Android
Application, (a) a sleep mode (No machine is ON), (b) all
machines is ON, (c) a virual operating machine of 20 inputs, and
(d) the setting functions. It was apparent that this application
could control all devices very accurately.
The authors are grateful to Research and Academic Service

Department, Thai-Nichi Instutute of Technology, for a partial
financial support. Grateful acknowledgements are also made for
Department of Industrial promotion, Ministry of Industry, for a
partial financial support.
REFERENCES
[1]
3. DISCUSSION AND FUTURE WORKS

This paper has presented an early work on IIoT as a proof-ofconcept. In comparison to the existing PLC system in
manufacturing process that licensed program and expertises are
required, this work has alternatively eliminated those licensed
program and the need in expertises through the use of
commercially avalibale Arduino micorcontroller with opensoftware. In addition, the wirelessly remote system has been
equiped for a feasiblity in coltrols. This could help decrease the
prices and human in manufacturing processes. In the future work,
this system will be tested in a real environment in terms of
variations ibn temperature and humidity. All devices will
integrated into a single one with design package obtained from the
3D printer. The I/O ports will be expaned to recive more input and
outputs, and test for its reliability.
CONCLUSION
A raodmap towards industrial 4.0 in Thailand and ASEAN
countries has aroused the need for Industrial Internet of Things
that enables remote communications via wireless IP devices and
interactions between human and manufacturing processes for the
enhancement of performances and efficiency as well as the
reduction in manufacturing costs. This paper has presented a
complete IIoT platform in which the universal PLC is controlled
by an open-software Arduino microcontroller via a high-voltage
peripheral board. The communication between PLC and end users
was achieved by Wi-Fi technology and MQTT internet protocol
that allow data storage and monitoring in cloud server. The GUI
was designed for both a web-based display and an Android
application. Demonstrations on the pneumatic control system
were also included. The proposed system offers a low-cost highefficiency implementation of IIoT for Small-and-Medium
Enterprises (SMEs).
Suranan
Noimanee
and
Somkiat
Wattanasirichaigoon,
Implementation of Vital Signs Monitoring System Using Wireless
Networks, International Journal of Applied Biomedical
Engineering, Vol.1, No.1, 2008.
In Lee, Kyoochun Lee, The Internet of Things (IoT): Applications,

investments, and challenges for enterprises, Business Horizons ,
vol. 58 , no. 4 , pp.431 440, 2015.
[3] Ciprian-Radu Rad aeorghe Olteanu, Smart Monitoring of Potato
Crop: A Cyber-Physical System Architecture Model in the Field of
Precision Agriculture, Agriculture and Agricultural Science
Procedia , vol. 6 , pp.73 79, 2015.
[4] Dan Koo, Kalyan Piratla, John Matthews, Towards Sustainable
Water Supply: Schematic Development of Big Data Collection
Using Internet of Things (IoT) , Procedia Engineering, vol. 118 ,
pp.489 497.
[5] Du Kun-kun, Wang Zhi-liang , Hong Mi, Human machine
interactive system on smart home of IoT, The Journal of China
Universities of Posts and Telecommunications , vol. 22, no.3, pp.96
99, 2015.
[6] Aditya Gaura, Bryan Scotney, Gerard Parr, Sally McClean, Smart
City Architecture and its Applications based on IoT, Procedia
Computer Science, vol. 52 , pp.1089 1094.
[7] Yuqi Wang and Hui Qi. Research of Intelligent Transportation
System Based on the Internet of Things Frame, Journal of Wireless
Engineering and Technology, pp. 160-166, Vol.3, 2012.
[8] Norbert Gal, Remote Ethernet PLC Networking and Control, The
proceeding of 17th Telecommunications forum TELFOR
2009,pp.759-762, 2009.
[9] P. Brizzi, D. Conzon, F. Pramudianto, M. Paralic, M. Jacobsen,
Claudio Pastrone, Riccardo Tomasi, Maurizio A., and Spirito, The
ebbits platform: leveraging on the Internet of Things to support meat
traceability, in a proceeding of EFITA2013, 2013.
[10] Sergey Popov and Evgeny Chernyy, Developing of energy
resources accounting and controlling system for the Internet of
Things, The proceeding of 14th conference of Fruct Association,
pp. 107 112, 2013.
[11] Felipe Corso, Yeison Camargo, and Leonardo Ramirez, Wireless
Sensor System According to the Concept of IoT -Internet of Things,
In the proceeding of International Conference on Computational
Science and Computational Intelligence, 2014.
[12] Mircea Murar1 and Stelian Brad, Monitoring and controlling of
smart equipments using Android compatible devices towards IoT
applications and services in manufacturing industry, in the
proceeding of IEEE International Conference on Automation,
Quality and Testing, Robotics, pp.1-8, 2014.
[2]
Watermelon Ripeness Extraction Based on Image

Color Segmentation and Rind Texture Analysis
Montri Phothisonothai*, Suchada Tantisatirapong**, and Apinan Aurasopon***
*
**
International College, King Mongkuts Institute of Technology Ladkrabang, Bangkok 10520, Thailand
Department of Biomedical Engineering, Srinakharinwirot University, Ongkharak, Nakhon Nayok 12110, Thailand
***
Department of Electrical Engineering, Mahasarakham University, Mahasarakham 44150, Thailand
E-mail: montri.ph@kmitl.ac.th, suchadat@g.swu.ac.th, apinan.a@msu.ac.th
AbstractWatermelon is mostly grown in all tropical areas of

agricultural countries especially in most of the Asian countries.
Quality control of the production is also needed to be a standardized procedure for an automatic system, e.g. computer vision
based system. In this paper, therefore, we objectively investigated the ripeness of watermelon based on color segmentation using
k-means clustering and rind texture analysis using Laplacian of
Gaussian (LoG) filter, which present a significant feature of its
ripeness level. Different experimental conditions were compared
to achieve the best performance. In our experiment, we captured
20 images of watermelon (Kinnaree variety) where 10 images are
selected quality in goods up to standard and 10 images are not
up to standard. The experimental results showed that the proposed features could extract different ripeness levels statistically
with p < 0.001.
KeywordsWatermelon Ripeness; Color Segmentation;
Nondestructive inspection; Image Processing; Edge Detection
In [7], the effective method to monitor of watermelon

ripeness by fuzzy logic has been proposed. The ripeness level
of the watermelon has been graded by human expert. However, this method requires expert based system training and
many input variable to obtain the best results. Recently, M.N.
Salim and Murinto [8] proposed statistic method whereas the
characteristic parameters are mean, variance, correlation, and
entropy as a method to identify the ripeness of watermelon
based on features of rind texture. However, this method presents inevitable weaknesses in terms of computational methods and limitation of rind image selection in specific region.
In this paper, we aim to determine whether watermelon is ripe
or unripe based on color segmentation using k-means clustering and rind texture analysis of automated region of interest
image classification.
II. MATERIAL AND PROPOSED METHOD
I. INTRODUCTION
Watermelon (Citrullus lanatus) is commonly grown and
consumed in all tropical areas of agricultural countries. For
example, in Japan, watermelons are popular and also have a
shape to provide value-added. In China, watermelons are the
top producer in the world. In Thailand, the country is in tropical weather which suitable for planting rice, fruit and vegetable, also the one of major food producer of the world each
year. Watermelon is also a tropical economic cucurbit crops
grown throughout the world. One of the main factors in ensuring the consistent marketing of watermelon is the quality of
the product. Undoubtedly, one of the most common problems
is that determining watermelon ripeness based on visual inspection. Traditional methods for assessing fruit ripeness are
destructive, thus cannot be so readily applied, particularly in
mass production [1].
For the related works, there were many non-destructive
methods proposed to determine the ripeness of fruits, e.g.,
durian [2], mango [3], potato [4], apple [5], orange [6] and
watermelon [7]. Therefore, it is important to monitor and
control the fruit ripeness since it is a major issue in fruit production.
Images of watermelon were acquired by using a digital

camera and 8-bit color resolution of 12807203 pixels in
bitmap (*.BMP) image format. The watermelons position is
placed approximately 100 centimeters away from camera as
depicted in Fig. 1. We set the A4 size of white paper sheets to
be a background.
Watermelon
100 cm
Digital camera
Input image
(*.BMP)
Fig. 1. Watermelons rind image acquisition in the experiment.
Ripe watermelon: ripped, moist and firm, red fleshy, and

sweet
Unripe watermelon: unripped, moist and firm, pinkishwhite fleshy, tasteless
All captured images in G1 and G2 datasets will be taken to
the proposed method as shown in Fig. 2. The proposed method consists of the following steps:
Input image
Normalization
Binary
Transformation
Morphological
Operation
1) Normalization: color image will be transformed into grayscale image. Due to the diversity of images in terms of contrasting or strengthening during acquisition process, the normalization is proposed to overcome that changes the range of
pixel intensity values to bring the image into a same range of
expected values, i.e., 0 255 in 8-bit resolution of this experiment. Normalization transforms an n-dimensional grayscale
image (n = 8) into the following equations:
Region of Interest
Classification
Feature Extraction
(
RGB Color
Segmentation
Rind Texture
Analysis
Fig. 2. System overview of the proposed framework, in this experiment, to

determine ripeness of watermelon by using image processing.
Background
)
(
(
)
(
(
)
)
(1)
(2)
where
(
) and
(
) are the minimal contrast function and maximal contrast function, respectively.
Output of this process is the image shading correction.
2) Binary Transformation: after the normalization process,
image will be transformed to a binary image based on the
histogram value thresholding. The histogram of image is
initially segmented into two parts using an average value of
(
) where
contrast value in each pixel, i.e.,
is a normalized contrast value ranged 0 1. Typical example of histogram based thresholding as shown in Fig. 3. The
binary transformation can be determined on the basis of the
following rule:
Object (watermelon)
T = 0.78
Fig. 3. The histogram of the grayscale image and the corresponding threshold
value to separate watermelon from its background.
In the experiment, we captured 20 images of watermelon

(Kinnaree variety) where 10 images are selected quality
(called G1 dataset used as a control group) in goods up to
standard and 10 images are not up to standard (called G2
dataset). By visual inspection, we subjectively defined the
meaning of ripe and unripe watermelons based on an experienced farmer and traditional knowledge as follows:
(3)
3) Morphological Operation: Morphological operation is a

technique of image processing based on shapes. The value of
each pixel in the normalized image is based on a comparison
of the corresponding pixel in the input image with its neighbors. This method is proposed to reduce the noise occurred
from the binary image, e.g., salt and pepper, and the shadow
of object. It presents itself as sparsely occurring white and
black pixels. The basic morphological operators are erosion,
dilation, opening and closing. In this paper, we choose the
opening operator to separate the object from background. The
opening of normalized image, , by the structuring element,
, is obtained by the erosion of by , followed by dilation

of the resulting image by
defined as
(
(4)
where
and are the structuring element of erosion and dilation, respectively. In our experiment, we set erosion
of 55
all-ones matrix and dilation
of 99 all-ones matrix.
4) Region of Interest Classification: this is called ROI classification, the output image can be obtained by the multiplication of image from opening operation. Therefore, the ROI of
image is a watermelon extracted from the background. For
examples, all results passed the proposed method can be seen
in Table I.
TABLE I. TYPICAL EXAMPLE OF THE OUTPUT IMAGES

PROCESSED BY THE PROPOSED FRAMEWORK.
Process
Result
Input Image
Normalization
Fig. 4. Expected values shown in vertical dotted line in RGB space of color
segmentation using k-means clustering where k is 3.
TABLE II. COMPARISON OF COLOR CLUSTERING FROM G1 AND

G2 DATASET.
Binary Transformation
Color Cluster
Morphological Operation
G1 Dataset (Ripe)
Pixel
[R, G, B]
42.33 %
[135, 138, 110]
33.44 %
[118, 120, 119]
24.23 %
[88, 89, 92]
G2 Dataset (Unripe)
ROI Classification
48.51 %
[97, 104, 87]
27.13 %
[123, 128, 120]
24.36 %
[81, 86, 81]
5) RGB Color Segmentation: Colors in the ROI image were

clustered into 3 groups by using k-means. The L*a*b* color
space (also known as CIELAB or CIE L*a*b*) is proposed to
segment the color from the CIE XYZ tristimulus values. The
L*a*b* space consists of a luminosity layer L*, chromaticity-layer a* indicating where color falls along the red-green
axis, and chromaticity-layer b* indicating where the color
falls along the blue-yellow axis [9].
(* p < 0.001)
Fig. 7. Feature extraction of the standard deviation (S.D.) and its moving
average was applied to determine the ripe and unripe data. The one-way
ANOVA was applied to test statistically significant among the groups.
All of the color information is in the a* and b* layers by

measuring the difference between two colors using the Euclidean distance in the RGB space as shown in Fig. 4. In order
to segment color by using k-means clustering, we used the
MATLAB function1 makecform('srgb2lab') in Statistics and
Machine Learning Toolbox. The experimental results of G1
and G2 dataset were summarized in Table II.
Fig. 5. Experimental results of rind texture analysis by using LoG based edge
detection corresponding to ripe watermelons.
6) Rind Texture Analysis: texture is characterized as the spatial distribution of the gray degree in a set of adjacent pixels.
Therefore, in this paper, we also investigate the rind texture of
watermelon since it considered as a unity depending on the
type of varieties. An image classification process is generally
requires the stage of feature extraction, which we proposed
the edge detection of the grayscale ROI image using zerocrossings after filtering with a Laplacian of Gaussian (LoG)
filter, i.e., MATLAB function2 edge(img,'log') in the Image
Processing Toolbox. The obtained results of rind texture analysis corresponding to ripe and unripe watermelons shown as
Figs. 5 and 6, respectively.
III. EXPERIMENTS AND RESULTS
Fig. 6. Experimental results of rind texture analysis by using LoG based edge
detection corresponding to unripe watermelons.
1)
Color-Based Segmentation Using k-Means Clustering,
2)
The experimental results are divided into two features, i.e.,

RGB color segmentation and rind texture analysis. For the
RGB color segmentation results shown in Table II, we have
investigated the ROI image by CIELAB color that shows the
highest color clustered in space green [135, 138, 110] and [97,
104, 87] for G1 and G2 datasets, respectively. Moreover,
these results indicate the rind texture color of watermelon
42.33% to 48.51%, i.e., trend to be more dark green and grey
lemon shading than unripe texture. The three color clustering
shows the different RGB color space which pixels of the
image assigned to each cluster. The border is the color of the
cluster as calculated by the average value of its pixels. However, we found that the color clustering information was sensitive to changing the expected value due to the experimental
Edge Image Processing [http://www.mathworks.com]
setting parameters, e.g., lighting, image resolution, position

of watermelon, rind texture orientation and contamination.
For the rind texture analysis results, as shown in Fig. 7,
the ROI images were processed by LoG based edge detection.
We compared the standard deviation (S.D.) and its moving
average of 300 pixels. The one-way analysis of variance
(ANOVA) was applied to determine statistically significant
among the groups. The experimental results showed that the
feature patterns of LoG based edge detection were effective
approach to extract the rind texture. We found that the rind
texture of unripe watermelon was significantly less complex
than ripe watermelon with p < 0.001. The reason is that the
rind texture related to the flesh and moist at various stages of
ripeness, i.e., unripped rind texture is glossy and thick [10].
This made the diffuse and specular reflection from its rind and
directly influenced the level of complexity in terms of ROI
selection. Accordingly, the obtained S.D. graphs as shown in
Fig.6, these present the significantly different among the ripe
and unripe datasets. The additional characteristic parameters
could enrich the image texture which eventually produces
more accurate decision [8].
IV. CONCLUSIONS
In this paper, we objectively determined the ripeness of
watermelon based on color segmentation using k-means clustering and rind texture analysis using Laplacian of Gaussian
(LoG) filter to extract the feature patterns. From this experiment, it can be concluded that the proposed method can provide a good possible combining of features in determining the
ripeness of watermelon non-destructively for the real-time
inspection system and to the mobile applications.
REFERENCES
[1] E. Llobet, E.L. Hines, J.W. Gardner and S. Franco, Nondestructive
banana ripeness determination using a neural network based electronic
nose, Meas. Sci. Technol. vol.10, no.5, pp.538-548. 1999.
[2] M. Phothisonothai, Nondestructive maturity classification of durian
based on fractal features, Int. Symp. Signal Processing and Its Applications (ISSPA), pp.566569, 2010.
[3] T. Charoenpong, K. Chamnongthai, P. Kamhom, and M. Krairiksh
Volume measurement of mango by using 2D ellipse model, IEEE Int.
Conf. Industrial Technology (ICIT), vol.3, pp.14381441, 2004.
[4] M. Barnes, T. Duckett, and G. Cielniak, Boosting minimalist classifiers
for blemish detection in potatoes, Int. Conf. Image and Vision Computing New Zealand, pp.397402, 2009.
[5] J. Zhou, H. Yin, J. Liu, and L. Fan, Method of image fusion for apple
surface quality detection, Automatic Control and Artificial Intelligence
(ACAI 2012), pp.13391342, 2012.
[6] C.P.D. Carolina and N.T.D. David, Classification of oranges by maturity, using image processing techniques, International Congress of
Engineering Mechatronics and Automation (CIIMA), pp.1-5, 2014.
[7] F. Y. A. Rahman and et al, Monitoring of watermelon ripeness based
on fuzzy logic, WRI World Congress on Computer Science and Information Engineering (CSIE '09), vol.06, pp.67-70, 2009.
[8] M.N. Salim and Murinto, Image processing application for detecting
the ripeness of watermelon based on features of the rind texture, Int.
Conf. Green World in Business and Technology (ICGWBT), 2014.
[9] C. Connolly and T. Fleiss, A study of efficiency and accuracy in the
transformation from RGB to CIELAB color space, IEEE Trans. Image
Processing, vol.6, pp.10461048, July 1997.
[10] Watermelon ripeness selection, wikiHow, http://th.wikihow.com (in
Thai)
ACKNOWLEDGEMENTS
This research was partially supported by the International
College (IC), King Mongkuts Institute of Technology Ladkrabang (KMITL). The authors would like to thank students
from Department of Electrical Engineering, Mahasarakham
University, who collected all images in the experiment.
Optimal Load Dispatch of Hydro Power Plant in

Central-1 of Laos Using Artificial Neural
Network
K. Vongsomphone, S. Premrudeepreechacharn And K. Ngamsanroaj
Faculty of Engineering, Chiang Mai University Chiang Mai, Thailand 50200.
khambai_vsp@hotmail.com, suttic@eng.cmu.ac.th. And Electricity Generating Authority of Thailand
Suratthani, Thailandkanchit.n@egat.co.th
he generation supplied to Electricite du Laos (EDL) grid is

with numbers of hydro power plants. Now the generation
supplied to the central-1 network of EDL is increasing.
Therefore, it is necessary to find the best economic way for
buying energy from other generating power plants at the
minimal electricity cost. The demand is growing more every
year. To ensure sufficient power supply nationwide. This paper
will present the optimal use of the hydro power plant in the
central-1 network of EDL by using artificial neural network
method.
There are several classical optimization techniques for
simulating the optimal power dispatch problem, such as lambda
iteration method, Newtons method [1-2]. Application of
artificial neural networks is based on real-time optimization of
thermal units. This method of economic dispatch of generating
units at different loads must be concluded to total fuel costs at
the minimum point [3-5]. MATLAB programming of different

methods has been used for solving economic load dispatch
problem. Lambda iteration method is implemented for three
cases of-without generation limits and losses, with generation
limits and without losses and generation limits and losses [4].
The economic dispatch of generating power using modified
lambda-method in practical situations and under normal
operating conditions, the generating capacity of power plants is
more than the total losses and load demand [6]. The method
based on lambda iterative approach using differential evolution
programming. The result of this research showed the economic
dispatch to minimize the operating fuel cost while satisfying the
load demand and operational constraints [7]. An approach of
economic dispatch in the practical hydrothermal power system
in Southern China [8]. The forecasting solar irradiance for solar
power plant by artificial neural network used backpropagation
algorithm [9]. Using neural network with backpropagation
learning algorithm for forecasting electricity demand [10]. The
neural network to optimize power dispatch method used less
time to computation the problem of optimization.
This paper proposes artificial neural network to find optimal
dispatching of hydro power generations in the central-1
network of Laos, which consists of the 5 hydro power plants for
finding the optimum cost.
The existing transmission system is the central-1 network of
EDL which covers the areas of Vientiane capital, the province
of Vientiane, Louangprabang, Xayaboury, Xaysomboun,
Xiengkhounag and Houaphan. This grid is supplied by hydro
power generations by EDL-Generation Public Company (EDLGen) namely: Nam Ngum 1 (NNG 1) with capacity installed
155 MW, Nam Leuk (NL) 60 MW, Nam Mang 3 (NM 3) 40
MW and Independent Power Producer for domestic (IPPd)
Nam Lik 1/2 (NL 1/2) with capacity installed 100 MW and Nam
Ngum 5 (NNG 5) 120 MW. The central-1 has been connected
The authors would like to thank the academic collaboration project between
the Electricite du Laos (EDL) and the Electricity Generating Authority of
Thailand (EGAT) for financial support for this research.
Khambai Vongsomphone (e-mail: khambai_vsp@hotmail.com) and
Suttichai Premrudeepreechacharn (e-mail: suttic@eng.cmu.ac.th) are with the

Faculty of Engineering, Chiang Mai University, Chiang Mai 50200 Thailand.
K. Ngamsanroaj (e-mail: kanchit.n@egat.co.th) Electricity Generating Authority
of Thailand Suratthani, Thailand.
Abstract This paper presents the optimal dispatch of the

hydro power plant in central-1 of Electricite du Laos (EDL) using
the backpropagation artificial neural network method. The
location of hydro power plant and distance are not at the same
location to supply the center of the load. The tariffs of generation
are different. The methods for optimal dispatch of hydro power
plants at different loads are presented. The objective is to minimize
the generating costs. This uses quadratic function costs for
generation, to find optimal dispatch of the hydro power plants by
using lambda-iteration method and artificial neural network
method. The testing results of the Lambda iteration method and
artificial neural network are compared with the existing actual
operating costs. The study results have shown that the optimal
operation cost is much lower than present plant operating costs.
Index Terms Optimal dispatch load, Hydro power plant,

Artificial neural network, Backpropagation.
I. INTRODUCTION
to the Electricity Generating Authority of Thailand (EGAT)

system with 115 kV transmission line [10-11]. There will be
additional generating connected to this existing system. This
makes the scheme of dispatching more complicated as shown
in Fig. 1.
mathematically described as follows [3-8].

CT
C1 C2 ... Cn
(1)
Ci ( PGi )
i 1
PGT
PG1
PG2 ... PGn
(2)
PGi
i 1
n
PGi
PD
(3)
i 1
MinPGi
Ci ( PGi )
MinPGi
i 1
PGi
PGi2 )
(4)
i 1
where:
n : The number of generator units,
i : Index of dispatch able units,
Ci : The cost generation of units i, [$/h]
CT : The total cost generation in system, [$/h]
PGi : The power generation of unit i, [MW]
PGT : The total power generation in system [MW]
i, i, i : Coefficients for power generation cost of unit i
Fig. 1. Single line diagram of the transmission system in central-1 network of

EDL.
A. Power balance
The optimal dispatch power generation of hydro power
plants should be such that total electric power generation equal
to the total of power transmission line loss plus total system
load demand.
That is:
n
(5)
PG P P
i
i 1
II. OPTIMAL DISPATCH PROBLEM

In this paper, a procedure of simulating coefficient of each
power generation is considered by using PowerWorld
simulation program and Excel function. First, the existing
power transmission system in Central-1 network of the EDL is
modelled in the Power World simulation program. Then, each
generator is modelled. After that, power flow and power losses
of power system when connected generators are calculated.
Finally, the cost function of each generator (i, i and i) is
analyzed by using Excel function as shown Fig. 2.
Existing power transmission system
Modelling power transmission system of
central-1 network of EDL by PowerWorld
Simulator
Modelling generator system
Analysis power flow and power losses
Analysis cost function of generation (i, i and i)
Fig. 2. Step of analysis coefficients of the quadratic generation cost function.
The objective of optimal dispatch is to define the generation

dispatch that minimizes the immediate operating cost, subject
to the constraint that total generation equal to total load plus
transmission losses. The economic dispatch problem can be
where: PD : Total load demand, and [MW]

PL : Power losses
[MW]
B. The inequality constraints
The generation limit of each unit has a minimum and
maximum.
(6)
PGi (min) PG PGi (max)
where:
PGi(min) : The minimum generation limit of unit i [MW],
PGi(max) : The maximum generation limit of unit i [MW]
III. OPTIMAL DISPATCH BY LAMBDA ITERATION METHOD
Figure 3 shows the block diagram for the step optimal
dispatch by the lambda-iteration method of solution for all of
generations, dispatching problem-neglecting losses. The
solution to this problem can be approached by considering a
graphical technique for solving the problem, then extending it
into the area of computer algorithms [1-3].
The algorithm for optimal dispatch of hydro power plant used
lambda iteration method. The program is calculated by using
MATLAB program. The first reads the system data, the cost
function of each generator (i, i and i), power limits of each
generation. The initial condition of lambda () that are constant.
They used to determine the power generation (PG1, PG2,
PG5) of each unit based on equal incremental cost. After
determining the generations for initial value of (). Looked out
differences value between total load demand and total
generating power from whole generators. To determine the
optimum generation scheduling is obtained or not obtained by
comparing with the agreeable tolerance (). If it has a different
value that is the agreeable tolerance. It will adjust a new () and

revert to starting to calculate the new generating power of the
generators. After obtaining optimum generation scheduling
then simulate the power generation cost by using equation (9).
Input
(X)
Neural
Network
(H)
Start
Set lambda
Adjust weights
Calculate PGi
For i= 1n
Compare
First iteration
Target
Fig. 4. The process of back propagation network.
Calculate error
=Pd - PGi
Adjust Lambda
Output
(Y)
Yes
No
No
Absolute error
tolerance
Yes
Print out lambda, power
generation & total cost
End
Fig. 3. Step of optimal dispatch by the lambda-iteration method.
The input unit Xi (i=1, 2, n) receives an input value then

broadcasts to hidden unit Hj (j=1, 2, n) Next, it will compute
its activation and sends result to output unit Y k (k=1, 2, n)
During training, each output unit compares its computed
activation with its target value to determine the associated error
for that pattern. This error is the base for calculating the factor
m (m=1, 2, n) it is used to distribute the error at output unit
Y back to all the previous hidden unit which are connected to
Y. It is also used to update the weights between the output and
hidden unit as shown in Fig. 4.
After the determination of factors, the weights of all layers
are adjusted at the same time. The adjustment to the weight
from hidden units to output units, is based on the factor and the
activation of the hidden units.
The equation of hidden unit
n
Hj
IV. NEURAL NETWORK METHOD
( X i wij ), j= 1, 2, . . . n
(7)
i 1
A. Backpropagation algorithm
The equation of output unit
The artificial network method has several formats and new

models, were successfully developed. Such as optimal dispatch
generated power. The backpropagation is regularly utilized in
neural networks; this basic method makes it adaptable to several
applications. The general nature of the back propagation
training method means that a multilayer, feedforward net
trained by backpropagation can be utilized to solve problems in
many areas [13].
The algorithm of backpropagation is part of the network
training. It is related to three stages such as: the feedforward of
the input training pattern, the backpropagation of the associated
error and the adjustment of the weights.
Yk
( H j v jk ), k= 1, 2, . . . n
(8)
j 1
B. Backpropagation neural network for optimal dispatch

The targeted neural networks have three layers, which are the
input layer, hidden layer and output layer. The input layer is the
number of total load demand, it has only one neuron. The
hidden layer has 20 neurons. The output layer is the number of
generating power for each generator. There are five generators
in the transmission system. So the output layer will have five
neurons. The network is fully connected. The output of each
neuron is connected to all neurons in the hidden layer threw a
weight value which is not shown in the Fig. 5.
PG1
V1-1
h1
W1
Input Generator Unit
PG2
Input load
h2
PLoad
PG3
W20
PG4
V20-5
h20
Input layer
Hidden layer
PG5
Output layer
Fig. 5. Artificial neural network structure for optimal dispatch.
The neural network algorithm, backpropagation is used for

training by MATLAB program.
II.
i PGi
(9)
PGi2
where: i, i and i are the coefficients of the quadratic

generation cost function.
PG is the output of power generation of the Lambda
iteration method (MW), and PGn is the output of power
generation of the Neural network method (MW).
Table I shows the generation limits and coefficients of the
power generation costs and function of the 5 hydro power
plants. The coefficients of the cost function can be found from
the step of analysis coefficients of the quadratic generation cost
function by the motive of the base tariff in the power purchase
agreement between EDL and owner of generation plus power
transmission system loss by using PowerWorld simulation
program and Excel function for calculating the coefficient for
power generation cost. as shown in Fig. 2.
TABLE I.
Hydro Power
Plant
TABLE II. POWER OUTPUT OF LAMBDA-ITERATION METHOD OF

SCHEDULING HYDRO POWER PLANT 5 UNITS
SIMULATION RESULTS
The cost function of each generator is a non-linear function

and it is much more difficult to solve this equation. So use the
Lambda iteration method for simulation by using MATLAB
program. This method of program can be described by (9).
Ci ( PGi )
A. Simulation & Result

The simulation and result of an optimal dispatch system are
analyzed which are based on a load demand from 100 MW to
480 MW, each step by 20 MW. The results simulated from the
Lambda iteration method and Neural networks as shown in
table II and table III, which illustrates the scheduling of the
hydro power plant when the generators are operated by
constraint limits. As seen from the simulated results, the Nam
Ngum 5 hydro power plant (PG5) and Nam Lik 1/2 hydro
power plant (PG4) are generators selected first, when the load
demand is low. Because the generation cost of both is the
lowest, by the motive of the base tariff plus power transmission
system loss is cheaper than other generators. When the load
demand increases, due to the limitation of each generator, the
higher cost of the generator such as Nam Mang 3 (PG3) and
Nam Ngum 1 (PG1) hydro power plant in respectively are
selected according to load demand, and the cost of the
generator. Nam Leuk hydro power plant (PG2) is scheduled for
the last, because this is the most expensive of all the generators.
COST FUNCTION GENERATOR OF 5 HYDRO POWER PLANTS

Coefficient for power generation
cost
($/h)
($/MWh) ($/MW2h)
No.
Load
(MW)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
100
120
140
160
180
200
220
240
260
280
300
320
340
360
380
400
420
440
460
480
Limit of generation
Min
(MW)
Max
(MW)
NNG1 (PG1)
442.39
54.109
0.0087
15.00
155.00
NL (PG2)
642.57
54.381
0.0200
20.00
60.00
NM 3 (PG3)
775.62
52.609
0.0198
10.00
40.00
NL 1/2 (PG4)
329.37
48.996
0.0154
20.00
100.00
NNG5 (PG5)
348.51
47.635
0.0204
20.00
120.00
Power output of the Lambda Iteration

Method (MW)
NNG1
NL
NM3
NL
NNG5
Cost
($/h)
(PG1)
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
20.00
40.00
60.00
60.00
74.21
88.20
102.19
116.17
130.16
144.14
160.00
(PG2)
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
20.00
25.79
31.80
37.82
43.83
49.84
55.86
60.00
(PG3)
0.00
0.00
0.00
0.00
0.00
0.00
0.00
20.00
40.00
40.00
40.00
40.00
40.00
40.00
40.00
40.00
40.00
40.00
40.00
40.00
(PG4)
39.05
50.42
61.79
73.17
84.54
95.91
100.00
100.00
100.00
100.00
100.00
100.00
100.00
100.00
100.00
100.00
100.00
100.00
100.00
100.00
(PG5)
60.95
69.58
78.21
86.84
95.46
104.09
120.00
120.00
120.00
120.00
120.00
120.00
120.00
120.00
120.00
120.00
120.00
120.00
120.00
120.00
7,457
8,465
9,479
10,500
11,529
12,564
13,609
14,669
15,746
16,831
17,922
19,019
20,120
21,225
22,336
23,452
24,572
25,697
26,827
27,962
TABLE III. POWER OUTPUT OF NEURAL NETWORK METHOD OF SCHEDULING

HYDRO POWER PLANT 5 UNITS
No.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
Load
(MW)
100
120
140
160
180
200
220
240
260
280
300
320
340
360
380
400
420
440
460
480
Power output of the Neural Network

Method (MW)
NNG1
(PGn1)
NL
(PGn2)
NM3
(PGn3)
NL
(PGn4)
NNG5
(PGn5)
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
20.00
40.00
60.00
60.00
74.21
88.20
102.19
116.17
130.16
144.14
160.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
20.00
25.79
31.80
37.82
43.83
49.84
55.86
60.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
20.00
40.00
40.00
40.00
40.00
40.00
40.00
40.00
40.00
40.00
40.00
40.00
40.00
39.05
50.42
61.79
73.17
84.54
95.91
100.00
100.00
100.00
100.00
100.00
100.00
100.00
100.00
100.00
100.00
100.00
100.00
100.00
100.00
60.95
69.58
78.21
86.84
95.46
104.09
120.00
120.00
120.00
120.00
120.00
120.00
120.00
120.00
120.00
120.00
120.00
120.00
120.00
120.00
Cost
($/h)
7,457
8,465
9,479
10,500
11,529
12,564
13,609
14,669
15,746
16,831
17,922
19,019
20,120
21,225
22,336
23,452
24,572
25,697
26,827
27,962
Fig. 6 shows the comparison power output between Lambda

iteration method and Neural network method, the optimal
dispatch generated power of each generator The simulation
result of both methods is very similar.
Total load
load. You can choose the generator or generators with the

lowest operating costs. Note that the costs are much higher. If
the generators are not switch at low load. When operating at full
load, the generation cost is no different. see Fig. 7.
Fig. 7. Comparison actual generation cost and optimize operate cost.
V. CONCLUSION
The This paper has presented the optimal dispatch of hydro
power plant scheduling. It provides the technical method to
optimally determine generation in the central-1 of EDL of Laos.
The objective function is to minimize of generation costs. This
paper proposes the neural network and lambda iteration
method, which is used for finding the optimal dispatch of the
hydro power generation. The simulation results of both methods
are very similar. However, the neural network method uses less
time than the lambda iteration method. When compared
between actual operation and the technical method. We can
observe that the optimize operation cost is lower than actually
operation cost.
ACKNOWLEDGMENT
PG1, PGn1
PG5, PGn5
PG3, PGn3
PG4, PGn4
The authors gratefully would like to give a big thank to the

academic collaboration project between the Electricite du Laos
(EDL) and the Electricity Generating Authority of Thailand
(EGAT) for financial support for this research.
PG2, PGn2
REFERENCES
[1]
[2]
[3]
Fig. 6. The result of optimal dispatch by using Lambda-iteration and Backpropagation neural network method.
The obtained result of the neural network method and the

lambda iteration method are very similar. But the neural
network method is better than in facility of solution finding
answer more.
Comparision between the actual operating cost and optimize
operated cost by using articial neural networks, the values are
different. The optimize operation cost is lower than actually
operation cost. When the generators are running with a low
[4]
[5]
A.J. Wood and B. Wollenberg, Power Generation, Operation and

Control 2nd , New York: John Wiley and Sons, Inc., pp. 29-88, 1996.
P. Sriyanyong, Electric Power System Analysis , Dept. of teacher
training in electrical engineering. King Mongkuts University of
Technology North Bangkok. pp. 29-88.
M. Djukanovic , M. Calovic , B. Milosevic and D.J. Sobajic, Neural-net
based real-time economic dispatch for thermal power pants , IEEE
Transactions on Energy Conversion, Volume: 11, No. 4, pp. 755-761,
December 1996.
D. Rahul, G. Nikita and S. Harsha, Economic Load Dispatch Problem
and Matlab Programming of Different Methods , International
Conference of Advance Research and Innovation, pp. 202-207, February
2014.
S. Panta, S. Premrudeepreechacharn, S. Nuchprayoon, C.
Dechthummarong, S. Janjornmanit and S. Yachiangkam, Optimal
Economic Dispatch for Power Generation Using Artificial Neural
Network , IEEE International Conference on IPCE 2007, pp. 1343-1348,
December 2007.
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
D. O. Dike, A. M. Izuchukwu, O. George, Economic Dispatch of

Generated Power Using Modified Lambda-Iteration Method, IOSR
journal of Electrical and Electronics Engineering, Volume: 7, pp. 49-54,
jul.-Aug 2013.
N. Visali, M.S. Reddy and M.S.K. Reddy, Economic load dispatch of
thermal power plants using evolution technique including transmission
losses , International Conference of Advances in Electrical Engineering
(ICAEE), pp. 1-5, Jane 2014.
Z. Jizhong, X. Xiaofu, C. Kwok, S. Bin and L. Cheng, An Approach of
Economic Dispatch in the Practical Hydrothermal Power System , in
Power and Energy Engineering Conference (APPEEC), pp. 1-4, March
2011.
S. Watetakarn and S. Premrudeepreechacharn, Forecasting of Solar
Irradiance for Solar Power Plant by Artificial Neural Network , IEEE
Smart Grid Technologies - Asia, pp. 1-5, November 2015.
V. Sackdara, S. Premrudeepreechacharn and K. Ngamsanroaj,
Electricity Demand Forecasting of Electricite Du Lao (EDL) Using
Neural Networks , IEEE Rigion 10 Conference, pp. 640-644, November
2010.
Electricite du Laos, Power Development Plan (PDP 2010-2020) ,
Vientiane, Lao Peoples Democratic Republic, August 2013.
Electricite du Laos (EDL), Annual Report 2014 , Vientiane, Lao PDR,
2014.
L. Fausett, Fundamentals of Neural Networks Architectures, Algorithms,
and Applications" prentice-Hall, pp. 290-333, 1994.
Reliability Assessment of the Electrical

Distribution System of 22 kV of the
Khuatinueng Substation in Luangprabang Lao
PDR
K. Silivanh, Student, S. Premrudeepreechacharn,
Faculty of Engineering, Chiang Mai University, Chiang Mai 50200, Thailand
K. Kanchit Ngamsanroa
Electricity Generating Authority of Thailand, Suratthani, Thailand
Abstract This paper presents a reliability assessment study
of the electrical distribution system in Luangprabang in 20152025 by using methodology the connecting of a 22 kV grid system
in each feeder, of the Khuatinueng substation. The installation
switching devices aim to maintain the distribution system and
reduce the impact of the power outage of each feeder. This study
is used DIgSILENT program for reliability electrical distribution
system analysis. The study results have shown that system
reliability has been improved after installing switching devices in
the feeders.
Keywords
Reliability
assessment;
reliability
improvement; investment.
I.
INTRODUCTION
nergy is strongly related to economic prosperity, state

organization, and private sector well-being, the reliable
source of energy depends on effective policies on the part of
electric utilities, which main factors driving the economy to
grow steadily. So electrical distribution system reliability
indices are an important measure of service, quality of a
utility. System reliability has a high effect on customer
satisfaction and utility cost, especially in large city areas that
are important for the economy and tourism. The power
interruption in these areas will result in very high outage cost.
The reliability improves of electrical distribution system is
indicated the service and supplied electric power to the
customer with efficient gives customers strongly believe in
energy consumption increase. Over the past year, statistical
This work was supported by the Electricite Du Laos (EDL) and
Electricity Generating Authority of Thailand (EGAT) for funding support for
this research.
K. Silivanh is M. Eng. student at Faculty of Engineering, Chiang Mai
University and an engineer at Electricite du Laos (EDL), Lao PDR (e-mail:
Khonesavanh_edl@hotmail.com)
S. Premrudeepreechacharn is with the Department of Electrical
Engineering, Chiang Mai University, Chiang Mai 50200 Thailand (e-mail:
suttic@eng.cmu.ac.th).
K. Ngamsanroaj is with Electricity Generating Authority of Thailand
(EGAT) Thailand (e-mail: Kanchit.n@egat.co.th)
electrical outage and interruption time in Luangprabang has

steadily increased as shown in Table I, which this problem
damages to the customer significantly. In addition, data of the
outage and average failure frequency indices of the electrical
distribution network system, and data for Khwuatinueng
substation as shown in Table II
TABLE I
STATISTIC OF OUTAGE IN THE EACH MONTH AND INTERRUPTION TIME OF THE
KHUATINUENG SUBSTATION, FOR YEAR 2015
Name
of
feeder
F1
F2
F3
F4
F5
Total
Month
1
0
0
0
0
1
1
0
1
1
0
0
2
0
1
0
0
1
2
0 0
5 9
1 0
0 1
1 1
7 11
9 10 11 12
0 3
6 2
0 0
1 4
2 4
9 13
0
1
1
7
7
16
0
0
0
1
0
1
0
1
0
0
0
1
0
1
0
1
1
3
0
0
0
0
0
0
Total
Interruption
outage
time (Hrs/yr)
(No/yr)
3
0:42
27
12:54
3
2:29
15
1:51
18
4:53
66
22:49
TABLE II
STATISTIC FAILURE RATE OF THE EACH FEEDER IN THE KHUATINUENG
SUBSTATION
Feeder
Failure Frequency
(1/a. km)
Number of outage
(Time)
Outage
duration
(min)
F1
F2
F3
F4
F5
0.90
2.76
0.84
0.66
7.62
3
27
3
15
18
25.2
752.4
137.4
90.6
371.8
Therefore, the Electricite du Laos (EDL) enterprise

Luangprabang branch is planned and developed to improve
the reliability of the electrical distribution system in provinces
since 2015-2025 [1]-[2]. The planning to improve the
reliability of electrical distribution system and analysis, outage
cost of customer in Luangprabang is based on reliability
index: SAIFI, SAIDI and outage cost of the customers. The
electric cost customer is assessed and compared with a loss of
customers due to power outage. The results of this analysis
can help in the decision to plan to maintain and improve the
reliability of the electrical distribution system.
RELIABILITY IMPROVEMENT
Data collection of each substation and each feeder

Data of distribution network system
Data of installed Transformer capacity
Record outage
Number of customer
The main objective of this study is to reliability

assessment, of the electrical distribution system for supporting
economic growth of Luangprabang, but based on reasonable
economic - society growth within and evaluate the reliability
of the distribution system. So this study is to improve
reliability on the electrical distribution system of 22 kV in the
Luanprabang, and by reduction of SAIFI and SAIDI and
CAIDI [3] - [5].
Modeling analysis by using

DIgSILENT Software
Improvement of distribution network system

- Installation of additional distribution
equipment
A. A development plan and the reliability improvement

Assessment after developing and improving reliability of
electrical distribution system in the area. This study, also
considered the outage cost of the customer, reliability
assessment of distribution system in Luangprabang.
Reliability improvement
B. Reliability Improvement methodology

The reliability improvement and analysis in the
distribution system will operate inspection, ability technically
of the electrical distribution system to lead the development
and improvement of the electrical system, according plan
development and improvement of the EDL Luangprabang
branch enterprise in the year 2015-2025. It operates on the
basis that energy consumption increasing, efficiency of work
and level reliability of the distribution system [1].
The system integration 22 kV network system of each
feeder and installation switching equipment, in the industry of
government and private regions, as a means that scholars have
taken a lot to develop and improve the reliability of electrical
distribution system with a more reliable, which can put a
number of power outages and lost energy decreases.
Reduction period average service sustainable options, or
reducing the average number of customers affected by service
sustained interruption and activities Interrupted [1].
Y
Summerize
Fig. 1. The flowchart of work.
III.
STUDY SYSTEM
The Khuatinueng substation received high voltage 115kV

from Nam-Ngum 1 HP, includes two unit of transformers
rated 115/22kV, 12.5x2 MWA and has five feeders. The
Khuatinueng substation is used to supply energy to business,
residential and, government office areas in Luangprabang
District, which is shown in Fig. 2 [1].
For installation of switching devices on each point of each
feeder is connected between feeders of distribution network
system for the Khuatinueng substation. The proposed location
of each devices are located as shown in Fig. 2.
G
Nam Ngum 1 HP
F3. 5
2x17.5 MW
4x45 MW
Y
Khuatinueng SS
Y
115/22 kV
Y
2x12.5 MW
F2. 3
Y
Y
F3. 4
F2
F3. 2
F3
F4
F5. 1
F5
To Xiengnguen District
F2. 2
F1. 1
To Chomphet District
To Khai District
F5. 2
F3. 1
F1. 2
F2. 1
Phonexai District
F3. 3
F1. 3
F1
The purpose of this study is to evaluate the reliability

indices and methodology of system improvement by using the
DIgSILENT Power Factory program. The theoretical strategy
was to the integration of a 22 kV grid system of the each
substation and add protective equipment to reduce the number
of customers affected by outages to a minimum and to
improve the stability of the power system. Description stage
analysis, reliability is shown in Fig. 1 [6] [7].
Choice of optimal result
F5. 3
Pak-ou District
To Viengkham District
F5. 4
Pakseng District
F4. 1
II.
Breaker
LBS existing (Close)
LBS existing (Open)
LBS is a new installation (Close)
LBS is a new installation (Open)
Fig. 2. Single Line Diagram of New LBS Installation Equipment and LBS
Existing of the Electrical Distribution system in Khuatinueng
Substation [1].
Statistic energy consumption of customer in 2015 and data

used in this study from being EDL of the Luangprabang
branch, which is shown in Table III [1] [2].
TABLE III
DATA OF THE KUATINUENG SUBSTATION 2015
Feeder
Load (MW)
Number of Customer
Line Length (km)
F1
F2
F3
F4
F5
4.81
3.15
3.51
1.56
0.42
4,930
11,324
6,078
2,253
6,647
15
199.5
14
11
153.6
In 2015 incident total outage of the Kuatinueng substation

of 66 and interruption times 22.49 hours which can be divided
of the natural disasters is 29%, tree 34%, failure of equipment
is 17%, accidents is 11%, and animal is 9%, which is shown in
Fig. 3. That causes of the animal, mostly, from birds and
snakes, which ticker dropout fuse, sometimes, the also
affected to the substation system.
Traffic
accident
11%
Animal
9%
natural
disasters
29%
equipmen
t failure
17%
Analysis of customer damage in the outage is most

important to state electricity du Lao enterprise, and important
issues need to be discussed to lead improved and resolution,
reliability of the system, and to make the electrical distribution
system supply are efficient and reliable. So, state electricity du
Lao enterprise Luangprabang branch, perform to plan for
developing and improves the reliability of electricity
distribution system in Luangprabang, to make numbers of
outage, lost during power outage, during 2015-2025.
Statistical power outage, failure rate of the system, the number
of customer totally and the number of customers affected
when the outage ca been seen as shown in Table IV [4].
TABLE IV
SUMMARY OF CUSTOMER INTERRUPTION, OUTAGE DURATION AND FAILURE
RATE OF THE SYSTEM
Feeder
F1
F2
F3
F4
F5
The EDL Luanprabang branch has planned to improve

reliability on the electrical distribution system into each region
to accommodate energy consumption that increases in the
future, since 2015 2025 [2] includes: industry, business,
residential areas, and government offices. Therefore,
the
reliability assessment of the electrical distribution system is
the most important to reduce outage cost of the customer, for
the year 2015 2025. The analysis of SAIFI, SAIDI and,
CAIDI is operated on DIgSILENT program.
For the adding switching device or LBS on the each
feeder, it is a methodology to plan, develop and improve the
reliability of electrical distribution system the LBS is a device
that can cut and connect circuit, while has loaded. Therefore,
the add a device protection in each feeder, it gives regions of
outage, number of customers, economy and industry areas
affected reduce, which the details of additions equipment in
each feeder are shown in Table V and Fig. 2.
TABLE V
CASE STUDY, ADDED EQUIPMENT PROTECTION IN THE EACH FEEDER OF
KUATINUENG SUBSTATION
Case 1 Add LBS, 3 Units, in the feeder 1.
Tree
34%
Fig. 3. The cause of the outage (%) of year 2015.
Customers
serviced
(Unit)
4,423
8,171
5,879
2,038
4,005
Kuatinuang substation for supporting economic - social

growth.
Customers
interruptions
(Unit)
11,044
2,319
38,970
5,958
2,020
IV.
Failure
rate
(1/a.km)
0.6
0.7
0.6
0.6
0.8
Outage
duration
(Hours/yr)
0.42
12.54
2.29
1.51
4.23
CASE STUDY
This paper considers in the planning and development to

improve reliability of the electrical distribution system by
installation of switching equipment in the each feeder of the
Case 2
Add LBS, 3 Units, in the feeder 2.
Case 3
Case 4
Add LBS, 1 Unit, in the feeder 4.
Case 5
For, case 1, 3 and 4, the supply of electricity to the

Luangprabang District, customers has mostly is category
business around 70%, small industry 5%, residential 15% and
office of the government 10%. So, reduce the number of
customers that affects during the outage reduce. Thus we have
to add switching devices or LBS into each point of the each
district to improve a reliability on the electrical distribution
system in areas such.
For, case 2, the supply of electricity to the two districts
namely: Luangprabang and Chomphet districts, customers
mostly categorizing residential around 75%, business 10%,
small industry 8% and office of the government 7%. So, to
reduce the number of customers that been affects during the
outage to reduce. Thus we have to add switching devices or
LBS into each point in the network of the each district to
improve reliability on the electrical distribution system in
areas such.
For, case 5, the supply of electricity to the three districts
namely: Phonxay, Pakseng and Pak Ou districts, customers
mostly category residential 80%, business 10%, small industry
7% and office of the government 3%. So, to reduce the
number of customers that been affects during the outage to
reduce. Thus we have to add switching devices or LBS into
each point of the each district to improve reliability on the
electrical distribution system in areas such.
The result of the study, before of the reliability
improvement of the electrical distribution system of each
feeder, of the Khuatinueng substation, by adding switching

devices in the each feeder, which is shown in Table VI.
TABLE VI
THE SAIFI AND SAIDI BEFORE ADDING THE SWITCHING DEVICES
SAIFI
SAIDI
CAIDI
Outage cost
Feeder
(No/year)
(Hrs/year)
(Hrs/No)
($/year)
0.0073
0.015
2.00
87,071.92
F1
F2
0.1900
0.380
2.00
33,331.44
F3
0.0121
0.024
2.00
74,574.08
52,858.88
F4
0.0087
0.017
2.00
F5
0.1580
V.
0.316
2.00
30,753.92
RESULTS OF THE STUDY
The reliability improvement of the electrical distribution

system in Luangprabang, after installing, switching devices in
each point of each feeder for the Khuatinueng substation was
calculated reliability indices of the distribution system. The
results are shown in Table VI indicated that installing
additional protective devices (LBS) can make the reliability
indices uniformly. Additionally, installation switching devices
in the each point of each feeder indicate, the difference
between the values of SAIFI, SAIDI and outage cost, before
and after improvement of the each feeder, as shown in Fig. 4.
And 5.
TABLE VII
THE SAIFI AND SAIDI, AFTER ADDING SWITCHING DEVICES
No.
SAIFI
(No/year)
SAIDI
(Hr/year)
CAIDI
(Hr/No)
Outage cost
($/year)
Case 1
Case 2
Case 3
Case 4
Case 5
0.0037
0.0744
0.0048
0.0072
0.0872
0.007
0.135
0.010
0.011
0.165
1.820
1.640
1.594
1.532
1.890
79,235.45
27,331.78
59,435.54
43,344.28
24,510.81
In case 1, after adding the switching device with three

units, SAIFI reduce by 0.0036 (No/year), SAIDI reduce by
0.008 (Hrs/year) and CAIDI reduce by 0.18 (Hrs/No), when
compared with before adding the switching device of the
system.
In case 2, after adding the switching device with five units,
SAIFI reduce by 0.0716 (No/yr), SAIDI reduce by 0.794
(Hr/yr) and CAIDI reduce by 0.193 (Hr/No), when compared
with before adding the switching device of the system.
In case 3, after adding the switching device with three
units, SAIFI reduce by 0.0073 (No/yr), SAIDI reduce by 0.014
In case 4, after adding the switching device with one unit,
In case 5, after adding the switching device with four units,
Installing additional the switching devices in each the point
of each feeder, showing the value of SAIFI and SAIDI has
changed significantly in each feeder. So, reliability assessment
of the distribution system, including the frequency and
duration of the interruption, we still consider the investment to
improve and develop the system in each case, and compare the
rate of return within with investment of each case, as shown in
Table VIII and Fig. 6.
TABLE VIII
THE INVESTMENT TO IMPROVE OUTAGE COST OF CUSTOMER DUE TO POWER
OUTAGE, AND COMPARE THE RATE OF RETURN OF EACH CASE
Outage cost
Outage cost
B/C ratio
before
after
No.
Investment
/year
improvement improvement
Case 1
136,561.60
87,071.92
79,235.45
11.1
Case 2
136,561.60
33,331.44
27,331.78
5.55
Case 3
227,602.67
74,574.08
59,435.54
4.92
Case 4
45,520.53
52,858.88
43,344.28
4.78
Case 5
182,082.13
30,753.92
24,510.81
29.16
Fig. 4. SAIFI index (Number/year).
Fig. 6. Total outage cost and investment.

Fig. 5. SAIDI index (Hour/year).
In case 1, total investment in installation LBS devices three
units in feeder 1 is $ 136,561.60, before adding LBS has

outage cost: 87,071.92 $/year, but after adding LBS has
outage cost reduce 79,235.45 $/year, which reduce is 7,836.47
$/year, which rate of return of case 1 is 11.1 years.
In case 2, total investment in installation LBS devices three
units is $ 136,561.60, before adding LBS has outage cost:
33,331.44 $/year, but after adding LBS has outage cost reduce
27,331.78. $/year, which reduce is 5,999.66 $/year, which rate
of return of case 2 is 5.55 years.
In case 3, total investment in installation LBS device five
units is $ 227,602.67, before adding LBS has outage cost:
59,435.54. $/year, which reduce is 15,138.54 $/year, which
rate of return of case 3 is 4.92 years.
In case 4, total investment in installation LBS devices one
unit is $ 45,520.53, before adding LBS has outage cost:
43,334.28. $/year, which reduce is 9,514.6 $/year, which rate
of return of case 4 is 4.78 years.
In case 5, total investment in installation LBS devices four
units is: $ 182,082.13, before adding LBS has outage cost:
24,510.81 $/year, which reduce is 6,243.11 $/year, which rate
of return of case 5 is 29.16 years
From the results have shown that lowest cost of improving
the electrical distribution system is in case 4 with minimal
investment, but the rate of return more than others case. Total
investment in case 4 is $ 45,520.53, before adding LBS has
outage cost: 52,858.88 $/year, but after adding LBS has
outage cost 43,334.28. $/year, which reduce is 9,514.6 $/year,
the rate of return of case 4 is within 4.78 years, when
compared with 5 cases. For the reliability index SAIFI is
0.0872 (interruptions/yr), SAIDI is 0.165 (Hour/year) and
outage cost is 43,344.28 ($/year).
VI.
CAIDI =
Outage cos t =
where
i i
[2]
[3]
[4]
[5]
[6]
[7]
(1)
where li = the failure rate,

Ni = number of customer interruptions,
NT = number of customers served.
SAIDI =
Sum of Customer Interruption Duration

=
Total Number of Customer Served
l N r
i ii
NT
IER Pj
(4)
tj
IER =
interrupted energy rate,

t j = outage duration,
Electricite Du Lao Luangprabang province branch, Annual Report

2014, Luangprabang Lao (PDR) December 2014.
Electric Du Laos, Annual Report 2014, Vientiane, Lao PDR,
December 2014.
T. Tirapong and S. Titti, Reliability Improvement of Distribution
System Using Reliability Centered Maintenance, IEEE T&D
Conference and exposition 2014 PES, pp. 1-5 year 2014.
S. Khounnouvong, S. Premrudeepreechacharn and K. Ngamsanroaj,
Reliability Improvement of the Electrical Distribution System of the
Thanaleng Substation in Vientiane Capital, Laos, IEEE North
American Power Symposium (NAPS), 2015 Oct, pp. 1-6 year 2015.
Nguyan Hoang Viet and Nguyen Van Ban, Automation and Reliability
Improvement for Distribution Feeder, IEEE Probabilistic Methods
Applied to Power System (PMAPS), 11th, pp. 609 614 year 2010
June.
P. K. Goswami; S. Chowdhury; S. P. Chonwdhury; Y.H. Song and J. K.
Das, Reliability Evaluation of Distribution System, IEEE University
Power Engineering Conference (UPEC), 42ad, pp 158-166 year 2007
sept.
C. Sutthapanun; P. Jirapong; P. Bunchoo; P. Thararak, Reliability
Assessment Tool for Radial and Loop Distribution Systems Using
DIgSILENT PowerFactory, IEEE Electrical Engineering/Electronics, ,
Computer, Telecommucactions and Information Technology (ECTICON), 2015 12th International Conference on, pp 1-6 year 2015 June.
The outage cost which is an important factor for the

improvement of reliability indies are calculated as follow
NT
(3)
i i
REFERENCE
[1]
APPENDIX
l N
i ii
Pj = power loss of outage.
CONCLUSIONS
Total Number of Customers Interruption

=
Total Number of Customers Served
l N r
l N
where ri = average outage duration
This paper investigates the reliability assessment of the

electrical distribution system in Luangprabang, reliability
indices of the distribution system. These results are studied
SAIFI, SAIDI, CAIDI and outage cost of customer
interruption time reduced, before and after adding switching
devices of the each feeder.
SAIFI =
Sum of Customer Interruption Duration

=
Total Number of Customer Interruption
(2)
Low-Complexity Lossless Image Compression

for Low-Performance Microcontroller
based on Color-Region Segment
Choawanat Samkampang
Daranee Hormdee*
Embedded System Research and Development Group,

Computer Engineering Department, Khon Kaen University
Khon Kaen, Thailand
E-mail: choawanat@gmail.com
Embedded System Research and Development Group,

Computer Engineering Department, Khon Kaen University
Khon Kaen, Thailand
E-mail: darhor@kku.ac.th
Abstract Image compression/decompression is rather hard

on an embedded system due to memory constraints. This paper
presents a low-complexity lossless image compression that can be
used to decode and reconstruct image on a low-performance
microcontroller. The proposed method can be done by dividing
image into regions based on colors, and grouping clustering pixels
of each color as one of these objects; Rectangle, Line or Point.
Then, all objects would be formed to reduce data redundancy. The
output of this procedure is presented as Color-Region of each color
by identifying color value first and following by object of that color
in each row.
The testing results showed that the proposed image
compression method can provide better compression for all cases.
The compression ratio varied from 19.08% - 95.73% with higher
rate for fewer total number of different colors. Additionally,
higher clustering pixels of the same color in Rectangle object yields
to higher compression ratio while Line and Point objects give less
rate respectively.
KeywordsImage Compression for Embedded System;
Image Compression base on Color Region;
I. INTRODUCTION
Due to its inexpensive cost, nowadays, a TFT LCD screen has
become a default device for user interfacing in an embedded system.
Supporting both text and colored graphic makes LCD display so user
friendly.
Displaying on LCD screen has been controlled by a microcontroller
system, so microcontroller must save all images in its within its internal
memory. Nonetheless, low level microcontrollers usually have low
capacity of memory, which might not be enough for saving all images
in case of large files. Therefore, the concept of low-complexity lossless
image compression was created to reduce the data size, and it should be
able to make image reconstruction with low-performance
microcontroller.
Data compression is very significant to multimedia data
compression. Currently, there are several methods of data compression
both in lossless and lossy. Lossless data compression such as Huffman
Coding, LWZ or Deflate, which can be used to measure frequency of
each alphabet appearing and replace with shorter data or replace phrase
appearing the second time with phrase appearing the first time in order
to reduce data size. These methods have been used with ZIP programs
for reducing the size of documents including lossless image in any
formats such as GIF [1] or PNG [2]. The downside of lossless data
compression is that a lot of memory is required due to dictionary table
created during decoding.
On the other hand, Lossy image compression is used to compress
image such as Joint Photographic Experts Group (JPEG, JPG) via
Discrete Cosine Transform (DCT) technique [3]. This DCT is be used
to convert the signal (spatial information) into numeric data (frequency
domain) so that the images information exists in a quantitative form
that can be manipulated for compression. However, inverse DCT
process is rather complicated to deal with.
From those reasons explained above, both traditional lossy and
lossless data compressions seems to be too complex on a low
performance microcontroller system.
This paper suggests a novel method of lossless image compression
which can be decoded on a conventional low performance embedded
system. The concept of this work is based on the characteristic of
images; clustering pixels often be same color, then reducing data
redundancy by grouping these same color pixels into objects;
Rectangle, Line and Point.
II. LITERATURE REVIEW

Various lossless data compression techniques have been
implemented for the past decade. This section is divided into 3 parts:
techniques on lossless data compression, compressed image file
formats and some related works on image compression on an
embedded system, respectively.
A. Lossless Data Compression Techniques

Run-length encoding (RLE) is a very simple form of lossless data
compression in which runs of data are stored as a single data value and
count, rather than as the original run. RLE were employed in the
transmission of television signals as 1967 [4]. RLE is most useful on
data that contains many such runs. For example, simple graphic images
such as icons.
Huffman code is commonly used for lossless data compression. An
algorithm developed by David A. Huffman while he was a Ph.D.
student at MIT, and published in the 1952 [5]. The algorithm derives
dictionary table from the estimated probability or frequency of
occurrence for each symbol, more common symbols are represented
using fewer bits.
LempelZivWelch (LZW) is a universal lossless data
compression algorithm created by Abraham Lempel, Jacob Ziv, and
*Corresponding Author
Terry Welch. It was published by Welch in 1984 [6] as an improved

implementation of the LZ78 algorithm published by Lempel and Ziv in
1978 [7]. The algorithm is simple to implement, and has the potential
for very high throughput. LZW algorithm is working by search for
repeated words or messages in string, save them in the table, and then
replace the words with pointer to table index. As storing the value of
pointer used less data size than storing whole word, the size of data is
reduced.
Deflate is a data compression algorithm and associated file format
that uses a combination of the LZ77 algorithm and Huffman coding. It
was originally defined by Phil Katz for version 2 of his PKZIP archiving
tool [8]. Deflate algorithm is achieved through two steps, first STEP 1s
matching and replacement of duplicate strings with pointers, second
STEP 1s replacing symbols with new, weighted symbols based on
frequency of use. Deflate Compression has been used prevalently in file
compression software including PNG compression.
B. Compressed Image File Formats

GIF (Graphic Interchange Format) [1] is a primary standard of the
data reduction used for communicating on the internet, which was
developed by CompuServe. GIF images are compressed using the LZW
lossless data compression technique to reduce the file size without
degrading the visual quality, in which the key feature of GIF is
transparent backgrounds of animated images. However, the limitation
of GIF is that it can contain only maximum of 256 colors.
PNG (Portable Network Graphic) [2] file format was created as a
free, open-source alternative to GIF. Basically, PNG uses Deflate
compression technique that is more efficient than GIF. Moreover, the
PNG file format supports eight-bit paletted images (with optional
transparency for all palette colors) and 24-bit truecolor (16 million
colors).
JPEG is a lossy form of image compression based on the discrete
cosine transform (DCT) [3]. This mathematical operation converts each
frame of the raw image from the spatial domain into the frequency
domain. A perceptual model based loosely on the human psychovisual
system discards high-frequency information, and the frequency-domain
is a convenient representation of the image because the high-frequency
coefficients, which contribute less to the overall picture than other
coefficients, are characteristically small-values with high
compressibility
Among various image file formats, JPEG and PNG provide top
compression ratio, however, these file formats have been compressed
with Discrete Cosine Transform and Deflate algorithm respectively,
and for decompressing those require rather high resource especially the
memory capacity and processor performance [12], which seem to be
not suitable with a low-performance microcontroller system.
C. Related Works
Three previous relevant research works are listed in this section.
First research was about an image compression based on regionsegment technique which also used in this paper. The other 2 studies
focused on an image compression for low capacity and complexity
devices using run-length encoding technique.
Li Zhe-lin, et al suggested the new method of reducing the size of
cartoon image and published in 2009 [9]. Each color in an original
image was classified into a region. By deleting data inside except the
contour, could save data space in each region as default value: RGB-XY with Freeman Chain Code, and then compressing the acquired data
with Lempel-Ziv, further yields to the smaller size of data.
Roman Slaby, et al suggested the image compression as a
monochrome for displaying on LCD with low capacity devices and
published in 2013 [10]. Run-length Encoding technique was used for
white and black pixels in order to decrease data complexity.
In 2014, Bojan Stankic, et al suggested new file format used for
images named Embedded Run-length Image Encoding (ERLE) [11].
This method has high compression rate with low complexity of
algorithm for embedded devices with limited resources. ERLE

employed run-length encoding technique with color palette table and
minimized used color palette by reducing number of bits used to
represent each pixel color. As the color palette has fewer entries, the
algorithm can use index in the minimized color palette to represent each
color with fewer bits. A smaller number of bits per pixel will result in
smaller image size.
Combining these previous works, this research introduces a simple
lossless image compression method base on run-length encoding data
compression technique, as mentioned in related work [10, 11] along
with color-region segment in order to reduce redundancy colors
indicated in the research [9]. With this scheme, the compressed image
can be easily decompressed back even on a microcontroller system.
III. IMAGE COMPRESSION

For the proposed method, image data redundancy can be reduced
by grouping clustering pixels of each color in an image into an object
(STEP 1). After getting objects in each color row, the row containing
minimum of 4 pixels has been selected (STEP 2), as any row containing
fewer than 4 pixels has more overhead data size than the amount of data
that can be reduced. Lastly, all of non-selected pixels in Step 2 is
appended to the result continuously (STEP 3). Flowchart of the
proposed image compression method shown in Fig. 2.
Rectangle
Line
Point
Row_Index = 0
Col_Index = 0
Row_Index = 0
Start_Col_Index = 5
End_Col_Index = 6
Rect_Height = 7
Row_Index = 0
Start_Col_Index = 2
End_Col_Index = 3
6
(a)
0
(b)
3
6
(c)
(d)
Fig. 1. (a) Original Image, (b) Region and Objects divided for Pink, (c) Yellow
and (d) Brown.
Start Col
End Col
Start Col
End Col
Height
Color
Categorize by color within the image
RESULT OF SEARCHING ROW-REGION IN ORIGINAL IMAGE

Objects (Column Index)
Point
Line
Rectangle
Pixel
Row
Pixel
Count
Index
Count
4?
Column
TABLE I.
Start
2
4
1
0
2
4
1
4
3
Write each color in RGB format
Sort by row (from Row Index min to max), then group

clustering pixels into objects (point/line/rectangle)
Pink
(#FF80
80)
Yellow
(#FFFF
00)
Counting pixel for each Row Index,

ignore those Pixel Count < 4
Brown
(#8050
00)
Write Row_index
Write all Point objects in

List of { Column } form.
Follow by End_of_point code (0ffh)
0
1
4
0
3
-
4
-
1+1
1+1
5
9
2+1
0
4
0
4
4
4
4+4
1
-
14+2+1
B. STEP 2: Selecting Row-Region and Forming Data
Write all Line objects in

List of {Start Column, End Column} form.
Follow by End_of_line code (0ffh)
Write all Rectangle objects in

List of {Start Column, End Column, Height}
form. Follow by End_of_rectangle code (0ffh)
Repeat for the next Row_index until complete for all rows.
Write End_of_color code (0ffh)
Repeat for the next color until complete for all colors
Selecting Row-Region process can begin, which only RowRegion containing less than 4 pixels will be rejected. The selected rows
was shown in the last column of Table I. Then, all selected rows are
saved in the data structure shown in Fig. 3.
The data structure of the proposed method starts with Image
Header and followed by Color-Region of the image (Fig. 3 (a)). Image
Header consists of Image_Width, Image_Height and Data_Length of
Color-Region (Fig. 3 (b)).
Color-Region is the region of each color within the image (Fig.
3 (c)). Color value is saved in RGB format, followed by Row-Region
data and end with Ending code of color, 0ffh (Fig. 3 (d)). Row-Region
data is listed with selected Row-Regions of the color in ascending order
of Row Index (Fig. 3 (e)).
Object is a group of clustering pixels found in each RowRegion, which can be divided into 3 types: Point, Line and Rectangle
(Fig. 3 (f)). All objects are saved to the list according to the type in
ascending order of Column Index, and following by Ending code of
object, 0ffh (Fig. 3 (g, i, k)).
For Point, Column_Index saved to list (Fig. 3 (l)).
For Line, Start_Column_Index and End_Column_Index are saved
to the list (Fig. 3 (j)).
Write colors for all ignored pixels in RGB

format from top right to bottom left.
For Rectangle, Start_Column_Index, End_Column_Index and

Rectangle_Height are saved to list. (Fig. 3 (g)).
To reduce the size of data saving, Row_Index and
Column_Index values are saved in 1 or 2 byte offset format. If the value
is less than 128, it can be saved as 1-byte format, if value is between
128 (28) to 16,384 (214), it will be saved as 2-bytes format which setting
most significant bit (MSB) as 1. However, the value over 16,384 (214)
is not supported here. Details are as shown in Table II.
End
Fig. 2. Flowchart of the proposed image compression method.
A. STEP 1: Searching Row-Region

Searching Row-Region can be started with categorizing an
image by colors, next considering each row, then grouping clustering
pixels of each color into an object which can be divided into 3 types:
Point, Line and Rectangle. A sample of searching Row-Region of
original image as Fig. 1 (a) consists of 3 color. After Categorize by color
and searching all objects, the result shows as Fig. 1 (b, c, d), and the
result of this step shown as Table I.
As example of Row-Region of pink shown as Fig. 1 (b),
considering first Row (Row_Index = 0) found that there are 3 objects
as Points on Column_Index at 0. Line on Start_Column_Index at 2,
End_Column_Index at 3 and Rectangle on Start_Column_Index at 5,
End_Column_Index at 6 with 7 pixels of Rectangle_Height.
Then, the same step was done with every Row-Region of every
color within the image.
TABLE II.
DATA FORMATION FOR ROW INDEX, COLUMN INDEX AND

RECTANGLE HEIGHT.
Condition
0
<= value < 128
128 <= value < 16,384
value >= 16,384
Data format
1 Byte Format, with setting MSB as 0
2 Bytes Format, with setting MSB as 1
Not supported
TABLE III.
Color-Region Data
(Step 1-2)
Image Header
Width
Height
Color #1
Color #...
Row Index
Color #...
Color #n
Selected Row-Region Data
Row-Region #1
Rect #...
Start_Column_Index
Rectangles
(f)
End of Rects
(g)
Rect_Height
(h)
End of Lines
(i)
End_Column_Index
(j)
Rect #n
End_Column_Index
Line #1
Line #...
Line #n
Start_Column_Index
Point #...
(d)
(e)
Lines
Rect #1
Point #1
End of Colors
(c)
Row-Region #n
Row-Region #...
Points
(a)
(b)
Color-Region Data length
Color #1
G
Non-selected pixels list

(Step 3)
End of Points
(k)
Column_Index
(l)
Point #n
Fig. 3. Data structure of the Proposed Method base on Color-Region Segment
C. STEP 3: Filling Non-Selected Pixels

By the end of the results from STEP 2, filling non-selected
pixels is applied with the color value in RGB of every pixel that had not
been marked, starting from the top-left to the bottom-right corner,
skipping the pixel that was marked. The result from this step can be seen
as in Fig. 4.
After completing all 3 compression steps, the result from STEP
2 is shown as Table III, and result from STEP 3 is shown as Table IV,
which all data of 83 bytes (include 8 bytes of image header).
0
0
1
2
Data Structure
Compressed Image
Image Header
Color #1 (Pink)
Row #0
List of Point
End of Point
List of Line

End of Line
List of Rectangle

End of Rectangle
End of Color
Color #2 (Yellow)
Row #2
List of Point
End of Point
List of Line

End of Line
List of Rectangle
End of Rectangle
Row #4
List of Point
End of Point
List of Line
End of Line
List of Rectangle

End of Rectangle
End of Color
Color #3 (Brown)
Row #3
List of Point
End of Point
List of Line
End of Line
List of Rectangle

End of Rectangle
End of Color
TABLE IV.
Fig. 4. Original Image after all pixels in selected rows have been marked X
Location
(Row, Column)
(0, 1)
(0, 4)
(1, 0)
(1, 1)
(1, 2)
(1, 3)
(1, 4)
(3, 1)
(3, 2)
(3, 3)
THE RESULT OF STEP 2

Data / Description
0x0007,
(Image_Width
= 7)
0x0007,
(Image_Height
= 7)
0x0000002D
(Data_Length
= 45)
0xFF8080
0x00
(Row = Row_Offset + 0 = 0+0 = 0)
0x00 (Col = Col_Offset + 0 = 0+0 = 0)
0xFF
0x02, (Start = Col_Offset + 0 = 0+2 = 2)
0x01
(End = Col_Offset + 1 = 2+1 = 3)
0xFF
0x01, (End = Col_Offset + 1 = 5+1 = 6)
0x07
(Rectangle_Height
= 7)
0xFF
0xFF
0xFFFF00
0x02
(Row = Row_Offset + 2 = 0+2 = 2)
0xFF
0x04
(End = Col_Offset + 4 = 0+4 = 4)
0xFF
0xFF
0x02
(Row = Row_Offset + 2 = 2+2 = 4)
0xFF
0xFF
0x02, (End = Col_Offset + 2 = 1+2 = 3)
0x03
(Rectangle_Height
= 3)
0xFF
0xFF
0x805000
0x03
(Row = Row_Offset + 3 = 0+3 = 3)
0xFF
0xFF
Rectangle #1
0x00, (End = Col_Offset + 0 = 0+0 = 0)
0x04
(Rectangle_Height
= 4)
0x04,
0x00,
0x04
0xFF
0xFF
Rectangle #2
(Start = Col_Offset + 4 = 0+4 = 4)
(End = Col_Offset + 0 = 4+0 = 4)
(Rectangle_Height
= 4)
THE LIST OF NON-SELECTED PIXELS DATA

Data (Color in RGB format)
0xFFFF00
0xFFFF00
0x805000
0x805000
0xFF8080
0x805000
0xFF8080
0xFF8080
0xFF8080
0xFF8080
(Yellow)
(Yellow)
(Brown)
(Brown)
(Pink)
(Brown)
(Pink)
(Pink)
(Pink)
(Pink)
IV. IMAGE RECONSTRUCTION
Function GetNextValue - Decode data base on condition from Table II.
The image reconstruction for drawing on LCD screen, compressed

image data can be decoded by the pseudo-code below, where
Compressed_Image_Array is array of compressed image data and
Read_Next_N_Bytes() function is applied for reading next value in
N bytes array continually then parse to integer in big-endian byte order.
Function Main Decode Compressed Image and draw on LCD Screen
Begin
Itterator_Array_Index = 0;
Image_Width = ReadNext_2_Bytes ();
Image_Height = ReadNext_2_Bytes ();
Data_Length = ReadNext_4_Bytes ();
/*---------- Decode Color-Region Data ---------- */
While (Itterator_Array_Index < Length)
Begin
Pen_Color = ReadNext_3_Bytes();
Row_Offset = 0;
While ((Row_Index = GetAbsoluteRowIndex()) != -1)
Begin
Col_Offset = 0;
// Draw Rectangles
While ((Start_Col_Index = GetAbsoluteColumnIndex()) != -1)
Begin
End_Col_Index = GetAbsoluteColumnIndex();
Rectangle_Height = GetNextValue();
LCD_DrawRect (Pen_Color, Row_Index,
Start_Col_Index, End_Col_Index, Rectangle_Height);
End While
Col_Offset = 0;
// Draw Lines
Begin
End_Col_Index = GetAbsoluteColumnIndex();
LCD_DrawLine (
Pen_Color, Row_Index, Start_Col_Index, End_Col_Index);
End While
Col_Offset = 0;
// Draw Points
Begin
LCD_DrawPixel(Pen_Color, Row_Index, Start_Col_Index)
End While
End While
End While
/*---------- Draw Non-Selected Pixels ----------*/
For (x = 0; x < Image_Width; x++)
Begin For
For (y = 0; y < Image_Height; y ++)
Begin For
If (Reconstructed Image at (x, y) have not been drawn)
Pen_Color = ReadNext_3_Bytes();
LCD_DrawPixel(Pen_Color, y, x);
End If
End For
End For
End Function
GetNextValue() is value transformation function according to the

condition in Table II. In addition, GetAbsoluteRowIndex() and
GetAbsoluteColumnIndex() are the row/column index value
transformation function from offset value to absolute value. These 3
functions can be shown by pseudo-code as follows;
Begin
d1 = Compressed_Image_Array[Iterator_Array_Index++];
If (d1 < 128)
Return d1;
Else
d2 = Compressed_Image_Array[Iterator_Array_Index++];
Return ((d1 - 128) * 256) + d2;
End If
End Function
Function GetAbsoluteColumnIndex - Get absolute value of Column Index
by adding with Column Offset
Begin
If (Compressed_Image_Array[Iterator_Array_Index] == 0xFF)
Itterator_Array_Index = Iterator_Array_Index + 1;
Return -1;
Else
Col_Offset = Col_Offset + GetNextValue();
Return Col_Offset;
End If
End Function
Function GetAbsoluteRowIndex - Get absolute value of Row Index by
adding with Row Offset
Begin
If (Compressed_Image_Array [Iterator_Array_Index] == 0xFF)
Itterator_Array_Index = Itterator_Array_Index + 1;
Return -1;
Else
Row_Offset = Row_Offset + GetNextValue();
Return Row_Offset;
End If
End Function
In
the
image
reconstruction
process,
the
Compressed_Image_Array has been read from start to end via
Iterator_Array_Index pointer without skipping or backwarding. So, the
algorithm complexity of reconstruction process is O(N) where N is
length of compressed image.
STEP 2 of the proposed method guarantees that the compressed
data size is not more than raw data size (N 3 x Image Size), so that
algorithm complexity is O(Image Size).
V. TEST RESULTS
The whole efficiency tests have been performed using STM32
CubeF7 Demonstration Application [13] on STM32F746G Discovery
Kit [14] with STM32F746NG microcontroller based on ARM CortexM7 core ( 1 Mbytes of flash memory and 340 Kbytes of RAM) with
4.3-inch 480x272 color LCD-TFT with capacitive touch screen (
Fig. 5). The main screen (Fig. 6 (a)) of demonstration application
consists with 8 image icons. Accessing each icon goes to various
screens (Fig. 6 (b-h)).
Each screen comprises of a number of drawing objects on a white
background. Those objects are divided into 3 types; polygon, text and
image (or icons). Polygons and texts could be drawn using sort of
various already built-in functions. However drawing images require all
image data to be stored in the internal memory as RGB format (3
bytes/pixels). All images in this test have been illustrated in red dashed
borders on each screen.
The testing started with measuring total pixel-count of all images
on each screen, later the number of clustering pixels on each screen
has been measured. Then, Grouping Ratio has be calculated. After
that, compassed data size and raw data size were measured. Finally
Compression Ratio has been evaluated. The testing results are as
listed in Table V.
The performance testing showed that compression ratio of the

proposed method compression depended on grouping ratio.
The result from Table V showed that the proposed method image
compression performed well for all tested cases here with compression
ratio varied from 19.08% - 95.73%. Let start by comparing between two
images with approximately the same raw data size (same image size):
Screen#2 and #8. As can be seen from the table that the compression
ratio of Screen#2 is a lot lower than Screen#8 that is because there were
so many total colors in Screen#2 with 5,163 different colors in it, while
only 713 colors in Screen#8. Screen#1 was the same story as Screen#8,
while the rest (Screen#3-7) had much higher compression rate due to
great clustering of solid colors.
From 8 experimental tests, the average of compression rate was
86.36%.
Fig. 5. STM32F746G-Discovery kit
VI. CONCLUSION
(a) Screen #1 - Main Screen
(b) Screen #2 - Audio Player
(c) Screen #3 - Video Player
(d) Screen #4 - Audio Recorder
This paper presented a low-complexity lossless image compression

that can be used to decode and reconstruct image on a low-performance
microcontroller. The testing results showed that the proposed image
compression method can provide better compression for all cases. The
compression ratio varied from 19.08% - 95.73% with higher rate for
fewer total number of different colors. An image with higher grouping
ratio can be compressed better than that the one with lower grouping
ratio. In addition, the compression ratio depends on how pixels of the
same color were clustered. Clustered in Rectangle object is better than
Line and Point respectively.
With a low-performance embedded system with only limited
memory on board, this image compression method should work just
fine although several image screens are required and still leave enough
memory for some internal onboard code.
REFERENCES
[1]
[2]
(f) Screen #6 - Home Alarm
(e) Screen #5 - Garden Control
[3]
[4]
[5]
(h) Screen #8 - System Info
(g) Screen #7 - VNC Server
[6]
Fig. 6. Tested Screens (tested images show in red dashed border)
[7]
TABLE V.
[8]
PERFORMANCE TESTING RESULT OF PROPOSED METHOD.

Pixel Count (Pixel)
Scre
en
All
Icons
#1
#2
#3
#4
#5
#6
#7
#8
Sum
29,335
19,650
24,310
44,400
47,880
51,402
62,400
19,600
298,977
Clusterin
g Pixel
20,924
4,404
21,711
40,601
46,527
47,991
58,789
13,767
254,714
a
Group
ing
Ratio a
(%)
71.33
22.41
89.30
91.44
97.14
93.36
94.21
70.24
85.20
Data Size (Byte)

Raw
Data
88,005
58,950
72,930
133,200
143,640
154,206
187,200
58,800
896,931
Proposed
Method
32,178
4,770
11,904
17,064
6,127
13,660
13,959
22,683
122,345
Compr
ession
Ratio b
(%)
63.44
19.08
83.68
87.19
95.73
91.14
92.54
61.42
86.36
Grouping ratio calculated by (Clustering pixel count / All pixels count).
Compression ratio calculated by (Raw data size Proposed method data size) / Raw data size.
[9]
[10]
[11]
[12]
[13]
[14]
CompuServe Incorporated. A standard defining a mechanism for the

storage and transmission of raster-based graphics information, 1987
M. Adler, T. Boutell, J. Bowler, C. Brunschen, A. M. Costello, L. Daniel
Crocker, et al., Portable Network Graphics (PNG) Specification (Second
Edition), 2003
International Telecommunication Unionm, T.81 Information technology
Digital compression and coding of continuous-tone still images
Requirements and guidelines, 1992
A. H. Robinson and C. Cherry, Results of a prototype television
bandwidth compression scheme, 1967, Proceedings of the IEEE, vol. 55,
issue 3, pp. 356-364
D. A. Huffman, A Method for the Construction of MinimumRedundancy Codes, in Proceedings of the IRE, 1952, pp. 1098-1101
W. Terry, "A Technique for High-Performance Data Compression",
Computer, vol. 17, issue 6, 1984, pp. 819
J. Ziv and A. Lempel, Compression of Individual Sequences via
Variable-Rate Coding, IEEE Transactions on Information Theory, vol.
24, 1978, pp. 530-536
L. Peter Deutsch, DEFLATE Compressed Data Format Specification
version 1.3, 1996
L. Zhe-lin, X. Qin-xiang, J. Li-jun and W. Shi-zi, Full Color Cartoon
Image Lossless Compression based on Region Segment, in Computer
Science and Information Engineering, vol. 6, 2009, pp. 545-548
R. Slaby, R. Hercik and Z. Machacek, Compression Methods For Image
Processing Implementation Into The Low Capacity Devices, in Tehniki
vjesnik 20, vol. 6, 2013, pp. 1087-1090
B. Stankic, D. Kojic, M. Cvetanovic, M. Dukic, S. Stojanovic, and Z.
Radivojevic, ERLE: Embedded Run Length image Encoding, 2014,
Telecommunications Forum Telfor (TELFOR), 2014 22nd , pp. 975-978
G. Roelofs, J. loup Gailly and M. Adler, zlib Technical Details, 2006,
STMicroelectronics, UM1906 User manual: STM32CubeF7
demonstration platform, 2015
STMicroelectronics, UM1907 User manual: Discovery kit for STM32F7
Series with STM32F746NG MCU, 2016
Remaining Lifetime Study of Generator

Step-Up Transformers in Thailand
Surin Sermsukroongsakul
Suttichai Premrudeepreechacharn
Department of Electrical Engineering

Chiang Mai University
Chiang Mai, Thailand 50200
surin.ser@egat.co.th
Department of Electrical Engineering

Chiang Mai University
Chiang Mai, Thailand 50200
suttic@eng.cmu.ac.th
Abstract - This paper study a remaining life of Generator

Step-Up (GSU) transformers in Electricity Generation
Authority of Thailand (EGAT). The remaining life of GSU
transformer depends on its solid insulation. Analysis of
insulation strength and statistical data are used to find a
lifetime curve model. By this lifetime assessment, it can
maintain availability and reliability of an electric
generation system. Also support Thailand Power
Development Plan (PDP) for extension lifetime of some
power plants in Thailand.
depend on its strength of insulating paper. The critical criteria

for remaining lifespan is the strength of insulating paper.
Degree of Polymerization (DP) of insulating paper is
commonly used for remaining lifespan prediction of power
transformer. A new insulating paper has DP value about
800~1000 [1]-[5]. The DP value of insulating paper will
decrease lifetime by factor of thermal, mechanical stress and
chemical reaction. The end of life for the transformer occurred
when DP value of insulating paper reaches 200 [2]-[8], the
paper becomes easily torn, and becoming less flexible.
Keywords - Generator Step-Up (GSU) transformer, Insulating Paper,

Degree of Polymerization (DP), Remaining life.
I. INTRODUCTION
An Electricity Generation Authority of Thailand (EGAT)
is a state-owned enterprise which has responsible to generate
an electric power in Thailand since 1969. Now EGAT has
installed capacity more than 16,000 Mega Watt (MW) with
131 units of Generator Step-Up (GSU) transformers in power
plants. GSU transformer step-up the voltage and transmitted
electric power from generator to power transmission line
system as shown in Fig. 1. Lifetime of power plants usually
depends on the lifetime of main equipment such as boiler,
turbine, generator and GSU transformer. Approximately 80%
of GSU transformers in EGAT were aged over 15 years, more
than half life of power plants. Asset Life Cycle Management
of main equipment such as GSU transformer is a key success
of power plant reliability and availability. The remaining life
assessment is an important tool for lifetime management of the
power plants in EGAT.
[Source: http://www.bu.edu]
Fig. 1. Electric power system.
The GSU transformer is one of the main equipments in

electric power plant with very high cost and can use it longer if
it is used and maintained properly, so it's not backed up with
complete set spare part. As shown in Fig. 2, most components
of the GSU transformer can be repair, replaced or recondition,
except for its solid insulation. While the tank, windings and
core will remain stable for the very long term. In normal aging
condition, remaining lifetime of GSU transformer mostly
The authors would like to thank The EGAT- CMU Academic Collaboration
Project between Electricity Generation Authority of Thailand (EGAT) and Chiang
Mai University (CMU) for technical and financial support for this research.
Surin Sermsukroongsakul (e-mail: surin.ser@egat.co.th) and Suttichai
Premrudeepreechacharn (e-mail: suttic@eng.cmu.ac.th) are with Department of
Electrical Engineering, Chiang Mai University, Chiang Mai 50200 Thailand.
[Source: http:// www.cst.com]
Fig. 2. Generator Step-Up Transformer.
II. DETERMINING LIFETIME OF POWER TRANSFORMER

A. Kraft Cellulose [2],[8]
Insulation in power transformer is consisted of cellulose in
feature of electric insulating paper, known as kraft paper.
Other polymer and pressboard are presented in kraft as well.
Kraft paper and pressboard have several functions in power
transformer like thermal, mechanical and electrical stability,
also conduct for oil flow direction.
Kraft paper is widely used for solid insulation in power
transformer due to its suitable in electrical and mechanical
properties, also good suit with mineral insulating oil as shown
in Fig. 3. The integration of mineral oil and kraft cellulose
provides better electrical properties. These electrical and
mechanical behaviors of kraft cellulose will change with ages
which depend on operating temperature, oxygen and water.
C. Aging of cellulose [2]

The molecular weight and DP value of insulating paper, are
constantly decreased during aging processes. That is the result of
cellulose chains are break into smaller chains during the
transformer operation with load. Aging of insulating paper can
indicated by chain cleavages, as describe in relation of chain
cleavage, and degree of polymerization as show in equation (2).
DP0
DPt
(2)
where DP0 is initial DP value

DPt is remaining DP value after time t
Chain cleavage value () is used to explain for accelerating
of aging process. The value of , it indicate the number of
extra small broken chains from cellulose chains.
The DP value of new kraft cellulose, DP0 is around
800~1000. After aging process, DP0 value is drop down to
DPt after duration time (t). DPt value can calculated by
Arrhenius equation as described by following equation [9]:
A t e
[Source: http://www.tbeaindia.com]
(3)
DPt
E
R (T 273)
1
DP0
Fig. 3. Solid insulation in power transformer.
B. Degree of Polymerization (DP) [2],[7]

Kraft paper in power transformer has a standard density of
~774 kg/m3 and contains moisture approximately 8%. Degree
of polymerization (DP) is defined as the number of repeat
monomeric units in an individual cellulose polymer chain as
shown in Fig. 4. DP value of kraft cellulose measure from the
kraft papers viscosity in special solution. The intrinsic
viscosity value is measured in accordance with ASTM
standard. The average degree of polymerization (DP) is the
ratio of monomeric molecular mass and average viscometric
molecular mass. This ratio related with intrinsic viscosity, as
the following equation:
[ ]
k DP a
where A
R
T
E
t
chemical environment factor

gas molar constant (8.3144598 J/mol/K)
absolute temperature (kelvin)
activation energy (111 kJ/mol)
duration (hour)
A tensile strength of cellulose at DP value 200, has

remaining ~25% of its initial value [8]. At a DP value less
than 300, the aging process decelerate, which is represent with
low gradient slope in Fig. 5. For insulating paper in
transformers, at DP value between 200 and 300 is usually
considered as the end of insulation service life. This range of
DP value is also defined as 'end of life' for power transformer.
(1)
where k = 0.0075
a = 1; for solvent polymer system
Fig. 4. Chain of cellulose polymer [11].
Lifetime of power transformer is principally determined by

thermal loading condition while insulation degradation is the
key of determining the remaining life of the transformer. New
insulating paper has a DP value approximately 1200 and
tensile strength 20 MN/m2. A DP value may decrease by 40
units during normal operating condition for 2 years.
Fig. 5. Degradation of DP value vs. time & temperature [8].
III. SAMPLINGS AND RESULTS
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
Insulating papers from inside of GSU transformers (as

shown in Fig. 6) could only collected a sampling during long
period shutdown of power plant. The sampling schedule needs
to synchronize with master shutdown plan of the power plan
from National Control Center (NCC) of Thailand.
(a) GSU transformer Overhaul Inspection
17.6
29.5
21.4
17.7
5.4
23.3
11.2
27.5
23.6
25.9
4.4
23.2
23.5
5.9
16.2
24.4
17.6
30.3
19.8
26.2
19.8
26.4
26.5
22.0
26.0
31.0
25.8
27.8
7.7
23.9
19.1
30.8
23.8
731
656
759
782
855
763
596
613
713
686
841
658
548
1028
766
629
939
679
676
567
775
454
596
654
734
720
717
785
959
898
729
792
807
From Table I, by plot of insulating papers strength (DP

value) and lifetime of GSU transformers. It shows an lifetime
curve for age estimation as Fig. 7.
(b) Opening top plate of GSU transformers

Fig. 6. Sampling insulation of GSU transformer.
Table I Shows Degree of Polymerization (DP) value of

insulating paper sampling from varied age of GSU transformers
from overhaul inspection between 2004 to 2015.
TABLE I. INSULATING PAPER: DP VALUE & GSU LIFETIME
Sample
Number
1
2
3
4
5
6
7
8
GSU
Transformer
Lifetime
(years)
15.6
21.9
14.7
25.9
5.5
5.4
21.9
23.3
DP Value
767
744
781
671
866
749
872
699
Fig. 7. Lifetime curve of GSU transformer in EGAT.
The estimated lifetime curve equation of GSU transformer in

years, EstLifeGSUTx(DP) is given by the following equation:
EstLifeGSUTx DP
35.26 264.131e
67592.46.DP-1.52
(4)
From the lifetime curve model as show in Fig. 7, its can

conclude that the strength or DP value of insulating paper
correlated significantly with operating time of GSU
transformer as equation (4) with correlation coefficient (r)
8.15. By this lifetime curve model, EGAT can estimate the
remaining life of GSU transformer in normal aging condition.
The reliability of an electric power transmission from
generator to power line system depends on the securely
operation of GSU transformer insulation system. Consider an
aging curve estimation of GSU transformer in Figure 7. At
service year of 30 when most of GSU transformers retired
with DP value more than 600. While the end of life considered
at DP value 200 for GSU transformer. This curve indicated that
most of GSU transformers still in good condition when it
retired. From this actual resulting of GSU transformer. It may
possible that manufacturers could consider to improve in
design criteria of insulation for reduce production costs.
IV. CONCLUSIONS
This paper has presented a lifetime study GSU transformer
in EGAT power plants for remaining life assessment. It
provides an aging model of GSU transformer. This aging
curve estimation is one of the most useful tools for lifetime
management of the power plants in EGAT. And also supports
the Thailand Power Development Plan (PDP) for extension
lifetime of power plants.
ACKNOWLEDGMENT
The authors gratefully would like to give a big thank to
The EGAT- CMU Academic Collaboration Project between
Electricity Generation Authority of Thailand (EGAT) and
Chiang Mai University (CMU) for technical and financial
support for this research.
REFERENCES
[1]
A.J. Kachler, I. Hhlein, Aging of Cellulose at Transformer Service

Temperatures. Part 1: Influence of Type of Oil and Air on the Degree of
Polymerization of Pressboard, Dissolved Gases, and Furanic
Compounds in Oil, IEEE Electrical Insulation Magazine, vol.21, no.2,
pp.15-21, 2005.
[2] Chaitanya Upadhyay, A study of life time management of power
transformers, Masters Thesis, E.ON Vrmekraft, resundsverket in
Malm, 2011.
[3] Hanung N.S., Suwarno, Nanang H., Yoshinobu Mizutani, Tsuguhiro
Takahashi, Tatsuki Okamoto, Degree of Polymerization Estimation of
Insulation Papers in Power Transformers Based on on Load and
Temperature Histories in Java-Bali Region of Indonesia National
Electric Company, IEEE International Conference on Condition
Monitoring and Diagnosis, 2012.
[4] Muhammad Arshad, Syed M. Islam, "Significance of Cellulose Power
Transformer Condition Assessment," IEEE Transactions on Dielectrics
and Electrical Insulation, vol.18, no.5, pp.1591-1598, 2011.
[5] Tetsuro Matsui, Yasuo Nakahara, Kazuo Nishiyama, Noboru Urabe,
Masayoshi Itoh, Development of Method for Remaining Life
Assessment of Oil-Immersed Transformer, Fuji Electric, 2010.
[6] A. Abu-Siada, S. Islam, "A New Technique to Estimate Furan
Concentration through Spectral Response of Transformer Oil,"
International conference of Doble Clients, Insulating Materials
Committee, 2014.
[7] S. Soares, N. Ricardo, F. Heatly, E. Rodrigues, "Low temperature
thermal degradation of cellulosic insulating paper in air and transformer
oil Society of Chemical Industry," Polym Int, vol.50, issue 3, pp.303308, 2001.
[8] C. Kuen, "Analysis and comparison of aging trends of cellulose for
transformer with oil-cellulose- insulation, IEEE International
Conference on High Voltage Engineering and Application, 2010.
[9] K. Najdenkoski, G. Rafajlovski, V. Dimcev, Thermal aging of
distributions transformers, IEEE-1-4244-1298-6/07, IEEE Power
Engineering Society General Meeting, 2007.
[10] Thomas A. Prevost, "Degradation of Cellulose Insulation in LiquidFilled Power Transformers", EHV-Weidmann Industries Inc., Fourth
Annual Technical Conference, 2005.
ICESIT 2016

Burapha University map
Entry

Ebook Icesit2016 Final PDF

Transféré par

Informations du document

Titre original

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

Ebook Icesit2016 Final PDF

Transféré par

Droits d'auteur :

Formats disponibles

ICESIT 2016

The 2016 International Conference on Embedded Systems and Intelligent Technology

August 5-8, 2016

Faculty of Engineering, 8 oor

Burapha University, Thailand

The 2016 International Conference on Embedded Systems

Session 1: Invited Talks #1-9,

Session 3: Keynote #1-3,

Session 4: Industrial Session

Session 2: Oral Presentation of

The 2016 International Conference on Embedded Systems

Session 5: Oral Presentation of

Session 2: Paper #6, 7, 8, 10, 12, 13, 16, 21, 24

Session 1: Room 1 (M801)

ICESIT2016 Page 1 of 119

The 2016 International Conference on Embedded Systems

Session 3: Room 1 (M801)

Session 2: Room 2 (M800)

ICESIT2016 Page 2 of 119

The 2016 International Conference on Embedded Systems

Session 5: Room 2 (M800)

ICESIT2016 Page 3 of 119

The 2016 International Conference on Embedded Systems

The 2016 International Conference on Embedded Systems

1. Sompol Pongthai (BUU, Thailand)

1. Prabhas Chongstitvatana (CU, Thailand)

2. Anat Deepatana (BUU, Thailand)

2. Kosin Chamnongthai (KMUTT, Thailand)

3. Apinetr Unakul (TESA, CAT Telecom)

4. Pansak Siriruchatapong (MICT)

1. Wiroon Sriborrirux (BUU, Thailand)

5. Narong Yoothanom (SPU, Thailand)

2. Takao Onoye (Osaka U, Japan)

6. Werachet Khan-Ngern (TESA & KMITL, Thailand)

3. Yo-Sung Ho (GIST, Korea)

8. Yoshikazu Miyanaga (Hokkaido, Japan)

Technical Program Chairs

19. Daranee Hormdee (KKU, Thailand)

2. Sanchai Eardprab (BUU, Thailand)

20. Watchara Chatwiriya (KMITL, Thailand)

3. Somjet Suppharangsan (BUU, Thailand)

21. Akkarit Sangpetch (KMITL, Thailand)

4. Anuparp Boonsongsrikul (BUU, Thailand)

22. Thumrongrat Amornraksa (KMUTT, Thailand)

5. Goparanjan Mohapatra (BUU, Thailand)

23. Wimol San-um (TNI, Thailand)

6. Norrarat Wattanamongkhol (BUU, Thailand)

24. Karn Opasjumruskit (Silicon Craft, Thailand)

7. Jantana Panyavaraporn (BUU, Thailand)

25. Nattha Jindapetch (PSU, Thailand)

8. Nayot Kurukitkoson (BUU, Thailand)

26. Supaporn Kiattisin (MU, Thailand)

9. Chalermpan Fongsamut (BUU, Thailand)

27. Preecha Kocharoen (SPU, Thailand)

10.Tharathorn boonsri (BUU, Thailand)

28. Sutthichai Premruedeepreechacharn

11. Jedsada Saai (BUU, Thailand)

12. Surachat Lekkham (BUU, Thailand)

29. Wimol San-Um (TNI, Thailand)

13. Supachai Vorapojpisut (TU, Thailand)

30. Virasit Imtawil (KKU, Thailand)

14. Wannarat Suntiamorntut (PSU, Thailand)

31. Somyot Wattanakamolchai (KMITL, Thailand)