Vous êtes sur la page 1sur 16

VEHICLE TRAJECTORY DETECTION USING MULTIPLE IMAGE TRACKING

July. 2009

Jutaek Oh The Korea Transport Institute

INTRODUCTION

VIPS provide traffic parameters in real time. However, most current VIPS have limitations in terms of safety. Do not identify individual vehicles as unique targets identif indi id al ehicles niq e targets. Do not trace vehicle trajectories. Detection zones are too short short. The objective of this research is to relate traffic safety to VIPS. Individual Vehicle Tracking Algorithm Long Distance Tracking Algorithm Incident Detection

Individual Vehicle Tracking Algorithm

Individual Vehicle Tracking Algorithm


1. Moving Object Segmentation

Background subtraction for extracting individual vehicle from road.

Individual Vehicle Tracking Algorithm


2. Determine the Threshold

1 if f ( x, y, t0 ) f ( x, y, ti ) > d 0 , j ( x, y ) = 0 otherwise
Where,

f ( x, y , t 0 )

Background template Current frame image 27

f ( x, y , t i )

Individual Vehicle Tracking Algorithm


3. Morphology

D( A, B) = A B =

U(A + )
B

: structuring element that determines the precise effect of the dilation on the input image.
The dilation Th dil ti morphology processing f extracting moving objects, and removing very h l i for t ti i bj t d i small particles as a noise.

Individual Vehicle Tracking Algorithm


4. Assign IDs
Reference Table ID Index X Y 1 0 250 380 Reference Table ID Index X Y 1 9999 250 380 2 0 360 380

A 1 2
When it comes to the zone, save ID to the reference table.

A 1 2

C(AB)

C(AB)

Multi Image Tracking Algorithm


Basic Concept
Used U d multiple cameras f l lti l for long di t distance t ki tracking. An intersection area between the overlapped images is unavoidable For example, the total surveillance area was around 200m with two cameras, the i t th intersection area was around 20 ti d 20m.

Multi Image Tracking Algorithm

DEMONSTRATIONS(I)

Jungang highway near by Wonju city in South Korea

DEMONSTRATIONS(II)

Another Experimental Site

Incident Detection : Back-moving vehicles

Incident Detection : Rapid lane changes

TEST RESULTS
Experiment has been conducted at Jungang highway in South Korea Compared with AUTOSCOPE (30 Minutes) Volume : 154 vs. 156 (98.7%), Speed : 54.8 vs. 54.5 (99.5%) 243 vehicles passed through the two zones (30 Minutes) Total of 236 and 241 vehicles passed through detection zones A and B, B respectively. ti l The error within the measured data was 2.07% inter-camera: 236 vs. 241 vehicles. Incident Detection Rate : 100% Back-moving incidents Rapid lane changes

CONCLUSIONS
Spatial traffic information can be more useful than spot information. Video image processing systems offer numerous benefits to transportation applications. The objective of this research is to develop VIPS using individual vehicle tracking and relate the developed system to traffic safety safety. Limitations of most current VIPS Traffic flow patterns cannot be estimated, if the detection zone is p short. Inefficient for incident detection The proposed tracking algorithm using multi image tracking will help to understand individual vehicle maneuvers and increase the effectiveness of incident detection detection. More than two cameras will be applied to an extended tracking distance of over 300~400m.

Thank You

Vous aimerez peut-être aussi