Vous êtes sur la page 1sur 5

2011 IEEE 22nd International Symposium on Personal, Indoor and Mobile Radio Communications

Fingerprinting-Based Radio Localization in Indoor Environments Using Multiple Wireless Technologies


Mois es Lisboa Rodrigues
Computer Science Department Univ. Fed. de Minas Gerais - Brazil moises@dcc.ufmg.br

Luiz Filipe M. Vieira


Computer Science Department Univ. Fed. de Minas Gerais - Brazil lfvieira@dcc.ufmg.br

Mario F. M. Campos
Computer Science Department Univ. Fed. de Minas Gerais - Brazil mario@dcc.ufmg.br

AbstractLocalizing a user is a fundamental problem that arises in many potential applications. The use of wireless technologies for locating a user has been a trend in recent years. Most existing approaches use RSSI to localize the user. In general, one of the several existing wireless standards such as ZigBee, Bluetooth or Wi-Fi, is chosen as the target standard. An interesting question that has practical implications is whether there is any benet in using more than one wireless technology to perform the localization. In this paper we present a study on the advantages and challenges of using multiple wireless technologies to perform localization in indoor environments. We use real ZigBee, Wi-Fi and Bluetooth compliant devices. In our study we analyse results obtained using the ngerprint method. The performance of each technology alone and the performance of the technologies combined are also investigated. We also analyse how the number of wireless devices used affects the quality of localization and show that, for all technologies, more beacons lead to less error. Finally, we show how interference among technologies may lead to lower localization accuracy.

I. I NTRODUCTION The problem of locating a user is a fundamental problem in many research areas. In outdoor environments, the Global Positioning System (GPS) can provide good location estimates. However, the GPS solution can not be used in indoor environments. In this kind of environment (which is typically called GPS denied environment) the GPS signal is very poor because of the lack of line of sight between satellites and the receiver. Due to the large number of applications that can benet from a location service in indoor environments, indoor location systems have been an important research topic in recent years. Because of the advances in wireless technologies and the consequent proliferation of wireless devices in indoor buildings, the use of radio frequency signals to perform localization has become an interesting and promising technique to build better location systems [1]. The idea is to use the received signal strength indication (RSSI) of packets sent by wireless devices available in the environment to perform localization. Most works in the eld of RSSI-based location in indoor environments use only a single technology [2], [3]. Since in many environments other technologies are available, it becomes attractive to use those technologies in location systems. In this paper we study localization in indoor environments using multiple wireless technologies. We consider three emerging wireless technologies: 802.15.4/ZigBee, Wi-Fi and

Bluetooth. We chose these technologies because of their growing availability1 . We use a mobile robot as the user, but our study is not limited to the case where robots are used. The user could be a human with a notebook or a smartphone, for example. In order to validate the methodology we performed experiments using the ngerprinting method. KNN (K-Nearest Neighbours) [2] algorithm is used to perform localization. The K-NN method was chosen because, despite its simplicity, it can achieve good results [5]. We analyse several aspects of localization using this method such as localization mean error and the inuence of the number of wireless devices used and the value of K (in K-NN algorithm) in the quality of the nal solution. We also show that, although combining the technologies could bring benets, the interference among them could compromise the localization performance. The main contributions of this paper that differentiate it from previous ones can be summarized as follows:

We use a combination of three current technologies, introducing Zigbee compliant devices, while other works are usually based just on Wi-Fi and/or Bluetooth or Wi-Fi and RFID. We show that by combining three technologies one can access better the potential and challenges of using multiple wireless technologies to perform localization. We introduce a merge process that can be used to fuse information from multiple wireless technologies when using the ngerprinting method. The merge process consists of concatenating each individual technology map. We show how interference among Wi-Fi and Bluetooth technologies can become a issue for localization accuracy. II. R ELATED W ORK

Most of the multiple wireless technology location works address the use of two specic technologies: WLAN and Bluetooth [6], [7]. Although, there are some works that use Wi-Fi and RFID [8], [9]. In those works, the location system is evaluated by comparing the behaviour of the system when using only one technology to the behaviour when using both
1 Although 802.15.4/Zigbee is not as highly available as Wi-Fi and Bluetooth in usual environments, it is becoming very popular in recent years and it is believed to be one of the pervasive wireless technologies of the future [4].

978-1-4577-1348-4/11/$26.00 2011 IEEE

1203

technologies. In what follows, we focus on the works that use a combination of Bluetooth and Wi-Fi. A method of sensor fusion between Bluetooth and Wi-Fi is presented in [6]. The method is divided into two phases. In the rst phase, the method uses the measures obtained from Bluetooth devices to select a zone where the object of interest is supposed to be. In the second phase, Wi-Fi signal measurements are used to locate the object considering only the region previously dened using the Bluetooth devices. When compared to the case that only uses Wi-Fi, the method combining Wi-Fi and Bluetooth estimates attained an average error 40cm smaller. According to the authors, the improvement on the localization quality when using multiple technologies is mainly due to the removal of outliers (the prior selection of regions eliminate outliers). A more comprehensive study of the effects of each technology on the quality of localization is presented in [10]. The experiments evaluate the impact of each technology on the quality of localization. They used ngerprinting for each technology. The inuence of different algorithms was also evaluated. Using Bluetooth achieves better results because of the more limited range of the technology, according to the authors. But, in some cases, the authors acknowledge that the results are not conclusive. [11] also uses Wi-Fi and Bluetooth to perform localization, but the focus of the work is not to make a study of the potential and challenges presented by the use of multiple wireless technologies. III. M ETHODOLOGY In this section we present an overview of the ngerprinting method and show how we combine the information of each technology to perform the localization. Then, we present the experimental setup and materials used. Finally, we discuss the data collection procedures. A. Fingerprinting There are two main methods to perform localization using radio (or radio location [1]): propagation model-based methods and ngerprinting-based methods. The methods based on propagation models use a mathematical model to predict the distance between sender and receiver based on the power with which a packet sent by a sender reaches a receiver. From this information, it is possible, by geometric computations (e.g. trilateration), to estimate the position of the receiver. The major drawback of propagation model-based methods is that the models used do not always hold in indoor environments due to phenomenons like multipath fading and shadowing [1]. This affects the accuracy of the localization process. In general, ngerprinting-based methods obtain better accuracy than those based on propagation model. This occurs because it avoids using propagation models that are not very accurate. We use ngerprinting method to localize the user. This method is separated into two phases. In the rst phase, called the training phase, a map of the environment is built. The map is a representation of the RSSI values at specic

training points of the environment. In the second phase, called the on-line testing phase, the user records the RSSI value at his location, and compares this value to values stored in the database created in the previous training phase. The nal user position estimate is the entry in the data base that best matches the actual measurements. In ngerprint-based methods there are deterministic and probabilistic techniques to perform the comparison step. In this work we use a deterministic method called K-NN. The basic idea of this technique is to compare the values collected in training and on-line phases by calculating the distance between them. This distance is calculated in the signal strength space. B. RSSI Map Building In order to use the method, we build a ngerprint map for each technology. The map consists of tuples [(xi , yi )(RSSI1 , RSSI2 , RSSI3 , ...RSSInt )], where (xi , yi ) represents a position in our grid and RSSIj , 1 j nt , t {Wi-Fi, Bluetooth, Zigbee}, is a RSSI reference value for each beacon. For each point of training, a tuple of RSSI values for each technology is collected. In the on-line test phase, the collected values are compared to the values stored in the database during the training phase. The K-NN algorithm returns, as the location estimation, the closest value of the database in relation to the tested point. We consider the euclidean distance to compare between trained and tested values (in signal strength space). For K = 1 the value returned by K-NN is the value of the closest training position. For K > 1, we calculate the mean of the coordinates (x, y ) of the K closest positions and take the mean as the nal location estimation. To combine the technologies, we merge the maps of each technology by a straightforward concatenation. For example, Figure 1 presents the nal result if we consider the case where we combine the three technologies.

Fig. 1. Merging maps from multiple technologies. The merging process consists of the concatenation of each individual technology map. The AP s, B s and Z s represents RSSI values from Wi-Fi, Bluetooth and Zigbee compliant devices, respectively.

C. Experimental Setup and Materials We use a Pioneer P3-AT mobile robot to perform the experiments. The robot was equipped with Bluetooth, ZigBee

1204

and Wi-Fi compliant devices, as shown in Figure 2.

Fig. 2.

Pioneer mobile robot with the sensing devices.

Fig. 3. Building area where we perform our experiments. The position of each wireless device and the training and test points are shown (this image is best viewed in color).

In order to collect data from all the wireless devices we used a Toshiba R Satellite M55-S331 notebook running Linux, Kernel 2.6.24, Ubuntu 8.04 - (Hardy Heron) distribution. The hardware conguration is an Intel R Pentium R M Processor 1.73GHz, with 495 MB RAM. Connected to the notebook was the USB-wireless adapter OvisLink AirLive R , the Billionton R Bluetooth USB dongle, and the USB programming board MIB 520 with a Zigbee radio (to collect data from sensor nodes). Six USB Bluetooth dongles, Noteship Micro Adaptor, by Leadership R were used as beacons. Five netbooks Eee PC, with Intel Atom N270 1.60 GHz (two cores) and 1GB RAM were used as base stations for the Bluetooth beacons. The last Bluetooth dongle was connected to a CCE Win notebook. We used eight micaz sensor nodes (as examples of 802.15.4/Zigbee compliant devices). The transmission power was congured at -7dBm. The Wi-Fi APs are from Cisco and we collect data from twelve of them (in Figure 3 there is just ve APs. We used other three on the same oor where the experiments were conducted and other four on the oor above the oor where the experiments took place). Our experiments were run in an area of 33 x 22 m2 . A map of the building area is shown in Figure 3. One of the sides of halls 1, 2 and 3 is half open 1.2 m above the oor. D. Data Collection We use technology-specic code to collect RSSI information. Data from Bluetooth devices were collected using the Linux Bluetooth library bluez. Sensor nodes data were acquired using the latest RSSIDemo application version [12]. To collect data from Wi-Fi APs, we use the iwlist Linux application. We now present how data were collected in training and on-line testing phase. We assume that environment remains unchanged from training phase to testing phase. Training phase. We use 49 training points. The training points are the little black squares shown in Figure 3. For each point we collected RSSI from Wi-Fi APs, Bluetooth

devices and ZigBee sensor nodes. The training points were separated 1.5 meters apart. For each position the robot stood still and collected 35 Wi-Fi signal samples, 30 from Zigbee (sensor nodes) and 30 from Bluetooth compliant devices for that position. Then, the robot was moved to the next position and repeated the data collection procedure. On-line testing phase. In the testing phase, we collected RSSI data at 15 points. Eight points were at the exact same location where training took place and seven points were at different locations from the training spots (see Figure 3). In the testing phase we collected the same number of samples that were collected in the training phase. IV. R ESULTS First, results comparing the accuracy achieved using various technologies combinations and values of K (the parameter of K-NN algorithm) are presented. Then, the robustness of the system is evaluated, i.e., we analyse the localization performance when locating the user in positions that were not previously trained. Finally, we analyse the inuence of the number of beacons used to perform the localization and show how the interference among the technologies could lead to lower localization performance. In the analysis, the mean error metric given by
em = 1 nt
nt

(xi x i )2 + ( y i y i )2
i=1

(1)

is used. In (1), (xi , yi ) represents the real robot position, ( xi , y i ) represents the estimated position, and nt is the number of tests performed. A. Mean Error Analysis Single technology case. Figure 4 shows the mean errors for each individual technology associated with different values of K. Wi-Fi achieves the best results and Bluetooth the worst. The success of Wi-Fi technology can be credited to the number of

1205

APs used. This number increases the probability of a location hit. Since there are many APs reference values for each position, the well known variability of RSSI has less impact on the results. This occurs because when the RSSI values of some specic AP has stronger variability between training and testing phase, the use of values obtained by other APs can reduce the error. In the case of the Bluetooth compliant devices and Zigbee sensor nodes, the communication range is reduced and this implies that fewer beacons (from the respective technologies) will be detected in each training position. Thus, if some of them have stronger variations, the impact on the nal result will be larger, since there are not many other values available that can be used to reduce the error. In relation to K values, for Wi-Fi, the error decreases until the value of K = 4 neighbours. After that, the error values increase again. This occurs because, for large values of K, the algorithm includes neighbours that are far away from the actual location where the robot is and this increases the nal location error. For Zigbee, we observed a similar trend, except that for K with values larger than 4 the error does not increase but remains constant. Bluetooth is almost K-invariant.

Fig. 5. Mean Error for different values of K, considering all test points and all technologies combinations.

at the chart for the Wi-Fi technology, we can realize that if the number of Wi-Fi APs available is approximately equal to the number of Zigbee and Bluetooth devices, the quality of location would not be better for Wi-Fi, as occurred in the mean error analysis presented in Section IV-A. This conrms the hypothesis that the success of Wi-Fi is associated to the number of available APs for this technology. In general, we can say that the larger the number of available beacons, the better the localization quality.

(a) Bluetooth

(b) Zigbee

Fig. 4. Mean Error for different values of K, considering all test points and each single technology.

Multiple technologies case. Figure 5 shows the results for each possible technologies combination. As expected, the cases where Wi-Fi is present, it is better than when it is not. Wi-Fi + Bluetooth and Wi-Fi + Zigbee have similar performances for K 4. For larger values of K, Wi-Fi + Zigbee outperforms Wi-Fi + Bluetooth. More specically, for values of K from 2 to 4, the error is about 1m, which is very reasonable, considering that the environment was not structured and the sensors used were not designed to obtain location information. Bluetooth + Zigbee performs better than each of them alone. Therefore, it is a good idea to use the combination if these are the only available technologies. Despite the relative poor performance of Bluetooth, the combination of all three technologies achieve good results, as shown by the black curve in Figure 5. B. Analysis of the inuence of the number of beacons In this analysis we investigated how the number of beacons used during the training and test phases affects the location quality. Figure 6 shows that, as expected, when the number of beacons used increases, the average localization error decreases. This occurs for all the three technologies. Looking

(c) Wi-Fi Fig. 6. Inuence of the number of beacons available in the localization quality. More beacons usually means less error.

C. System Robustness Analysis We also evaluated the robustness of the system by testing at locations that were not previously trained (see Figure 3). Figure 7 shows the mean error for all possible technologies combinations and values of K. The resulting errors in untrained positions is comparable to that obtained in trained positions. This suggests that it is possible to use the method to localize a user in a position not previously trained. This is important, since in real life applications the user seldomly is going to be exactly on the training positions. D. Analysis of the Interference Among Technologies and its Impact on Localization Accuracy The three technologies used in this work operate in the frequency of 2.4 GHz. Thus, interference could happen. Figure

1206

(a) Wi-Fi

(b) Bluetooth

(c) Zigbee

(d) Wi-Fi + Bluetooth

Fig. 8. Interference among Bluetooth and Wi-Fi technologies. The number of Wi-Fi APs detected decreases because of Bluetooth interference. Less APs detected could lead to lower localization accuracy (see Figure 6).

(e) Wi-Fi + Zigbee

(f) Zigbee + Bluetooth

accuracy can be improved by combining the technologies. The main challenge is to avoid adverse effects that arise when using multiple technologies (e.g., interference). These effects can reduce the quality of the localization. In future works, we plan to investigate in further detail how the interference among the technologies affects localization accuracy. ACKNOWLEDGMENT We are thankful to Tiago A. Arruda, Marlon L. Rodrigues and Wolmar Pimenta for their help with the experiments. Many thanks to Anderson Pires, Armando Neto, and the anonymous referees for their insightful comments on the text. We also thank CAPES, CNPq and FAPEMIG for the nancial support. R EFERENCES
[1] H. Liu, H. Darabi, P. Banerjee, and J. Liu, Survey of Wireless Indoor Positioning Techniques and Systems, IEEE Trans. on Syst., Man, and Cybernetics, vol. 37, no. 6, pp. 10671080, 2007. [2] P. Bahl and V. Padmanabhan, RADAR: An In-Building RF-Based User Location and Tracking System, in INFOCOM 2000, pp. 775784. [3] A. Haeberlen, E. Flannery, A. M. Ladd, A. Rudys, D. S. Wallach, and L. E. Kavraki, Practical Robust Localization Over Large-Scale 802.11 Wireless Networks, in MobiCom 2004, pp. 7084. [4] MIT, 10 Emerging Technologies That Will Change the World, Technology Review, pp. 3349, 2003. [5] X. Wu, V. Kumar, J. Ross Quinlan, J. Ghosh, Q. Yang, H. Motoda, G. McLachlan, A. Ng, B. Liu, P. Yu, Z.-H. Zhou, M. Steinbach, D. Hand, and D. Steinberg, Top 10 Algorithms in Data Mining, Knowledge and Information Systems, vol. 14, pp. 137, 2008. [6] S. Aparicio, J. Perez, A. Bernardos, and J. Casar, A Fusion Method Based on Bluetooth and WLAN Technologies for Indoor Location, in MFI 2008, pp. 487491. [7] S. Aparicio, J. Prez, P. Tarro, A. Bernardos, and J. Casar, An Indoor Location Method Based on a Fusion Map Using Bluetooth and WLAN Technologies, in International Symposium on Distributed Computing and Articial Intelligence (DCAI 2008). [8] A. Papapostolou and H. Chaouchi, Exploiting Multi-Modality and Diversity for Localization Enhancement: Wi-Fi & RFID Usecase, in PIMRC 2009, pp. 19031907. [9] , Simulation-Based Analysis for a Heterogeneous Indoor Localization Scheme, in CCNC 2010, 2010, pp. 15. [10] D. Pandya, R. Jain, and E. Lupu, Indoor Location Estimation Using Multiple Wireless Technologies, in PIMRC 2003, pp. 22082212. [11] A. Mahtab Hossain, H. N. Van, Y. Jin, and W.-S. Soh, Indoor Localization Using Multiple Wireless Technologies, in MASS 2007, pp. 18. [12] RSSIDemo, TinyOS RSSI Demo Application Website, Website, 2011, http://docs.tinyos.net/index.php/Rssi Demo. Last Accessed: April/2011.

(g) Wi-Fi + Zigbee + Bluetooth Fig. 7. Mean error in trained and not-trained positions. Errors in positions that were not trained are just a little larger than errors in positions that were trained. This is a desirable characteristic, since the user usually is not exactly on one of the training positions when using the method to locate itself.

8 shows an example of the problems that can arise because of interference. We performed a simple experiment where we collected the number of different Wi-Fi APs that our hardware could detect. In the rst phase of the experiment (rst third of the graph in Figure 8) we did not collect data from our Bluetooth compliant device. In the second phase, Bluetooth data collection starts. The number of APs detected immediately decreases, as shown in the second third of graph. Finally we stop the Bluetooth data collection and the number of APs the network adapter can listen to increases again. As our previous analysis has shown, the number of beacons used (detected, in this case) has great impact on the quality of the localization achieved. Therefore, interference among technologies can lead to lower localization accuracy. In the results previously presented, we did not consider the impact of interference among the technologies. But, as this experiment has shown, this is a relevant issue and further investigation is needed in order to better characterize this phenomenon and develop techniques that could deal with it successfully. V. C ONCLUSION AND F UTURE W ORKS In this work we evaluated the use of multiple wireless technologies to locate a user in indoor environments. The use of multiple wireless technologies to perform localization has advantages and challenges. The main advantage is that location

1207

Vous aimerez peut-être aussi