Dissertation > Transportation > Road transport > Technical management of traffic engineering and road transport > Computer applications in road transport and highway projects

Study and Application of Key Algorithm in Vehicle Visual Tracking

Author ChenShengLan
Tutor LongYongHong
School Hunan University
Course Control Theory and Control Engineering
Keywords vehicle tracking mean shift Kalman filtering adaptivekernel-bandwidth
CLC U495
Type Master's thesis
Year 2013
Downloads 22
Quotes 0
Download Dissertation

Intelligent transportation system is a main measure to solve the trafficcongestion, improve the traffic safety and protect environment. Based onthe analysis of the traffic video monitoring, the positioning and tracking ofvehicles can obtain the traffic information such as traffic flow, speed, roadoccupancy rate, queue length dynamic. Mean shift target is an algorithmwidely used in vehicle tracking using color as the tracking feature, canovercome the problems during tracking such as the light changes, partiallyoccluding, deformation, fast operation. However, it is very difficult as alocal optimization algorithm to deal with the situations that vehicle movestoo fast, moves under occlusion, lost temporarily, and the appearance of thevehicle changes. In this paper, according to above problems, the followingresearch work was made:(1) Mean-Shift algorithm combined with Kalman-filteringConsidering that the mean shift algorithm is a local optimizationalgorithm, and can not deal with the fast moving vehicles or vehiclesencounter the severe occlusion, we established the Kalman filtering vehiclemodel, put the results of mean shift tracking worked as a detection valueinto the Kalman prediction framework. We proposed the judgment methodfor severe occlusion, designed occlusion function judgment based on theBhattacharyya coefficient, expand the search area around the originalsearch point after the occlusion, which reduced the prediction error.(2) Adaptive kernel bandwidth of mean shift trackingConsidering when the vehicle becomes larger, there is deviation ofvehicle center and tracking window center, backward tracking combinedwith matching method based on the largest Bhattacharyya coefficient ofplus or minus ten percent window size is proposed. The algorithm takes less time, and can accurately locate the centroid, and can overcome theproblems of the window size getting larger by the method of ten percentwindow size correction. It can reflect the change of window size byexperimental verification.(3) the traffic information collection system based on mean shiftalgorithm is established by Matlab software. The software structure, thefunction and interface and the basic realization of algorithm acquisition isdesigned in the paper. The system can carry on the good positioning ofvehicle tracking, the fast tracking of targets and overcome the seriousocclusion, and has wide adaptive tracking window.

Related Dissertations
More Dissertations