Vehicle Detecttion And Tracking For Traffic Intersection

This paper establishes a vehicle detecting and tracking system for trafficintersection based on intelligent transportation system. With the increase ofmotor vehicles, deaths caused by traffic accidents have increased dramatically.Intelligent transportation system is an effective traffic management method.With the development of computer vision technology, road video surveillanceis becoming an important constituent of the intelligent traffic systemrealization by its offsite commanding,24-hour monitoring and quickjudgment of abnormal situation.Traffic intersections are complex locations where accidents and trafficviolations occur frequently. The study of vehicle detection and tracking fortraffic intersection has great significance. The characteristics of the trafficintersection include “heavy traffic” and “intensive vehicles”. The videosequence acquired is usually close-range video. So the research difficultiesare how to detect and segment vehicles accurately and how to overcome significant scale change of vehicle size.This paper designed a vehicle detection system and vehicle trackingsystem for traffic intersection video sequence and solved the two problemsmentioned above.For vehicle detection system, algorithms based on background modelingfit for the stationary camera. Combining the characteristics of trafficintersection, we proposed an effective segmentation method. We project theforeground detection map and segment vehicles by detecting peaks andextreme points of projection curve.Vehicle tracking system is based on Mean Shift algorithm. However,Mean Shift is not robust when scale of vehicles change significantly, so itcannot be used for long range vehicle tracking. We proposed an improvedvehicle tracking algorithm based on inverse perspective transform. Withplanar homographies, we reconstruct another view from original images. Theexperiments show that the proposed algorithm is robust and effective indealing with the traffic intersection video sequence. Combined with thevehicle detection system we achieved automatic detection of vehicles andlong-range stable tracking system for video sequences with a large trafficflow.