Study On The Video-based Traffic Detection And Tracking And Behavior Analysis Of Moving Vehicles

With the rapid development of economy, the number of vehicles is increasingdramatically, while the road traffic condition is deteriorating, so the traffic congestionbecomes more and more serious. In order to solve the traffic problems, the research of theintelligent transportation system has attracted more and more attention. Automaticsurveillance system is an important part of the intelligent transportation system. Detectionand tracking technique for moving vehicles in video sequences and trajectory analysis is thekey technology of automatic monitoring system. The research in this technology not onlymakes the intelligence level of the traffic monitoring greatly improved, but also show animportant significance in reducing traffic congestion, ensuring the fluency and the safety ofthe traffic and improving traffic management efficiency. On the basis of the generaltechnologies of vehicle detection and tracking, the paper emphasizes the moving targetsdetection and extraction, accurately tracking and trajectory analysis method. The thesismainly focused on subjects as follows:(1) In extracting the model of the target, connected component labeling algorithmcommonly used the pointer structures like linked list, tree and stack. According to theproblem that recursive invocation and pointer transmission cost much time by the structures,an optimized connected component labeling algorithm was adopted to label the connectedregion of differencing images obtained by using frame difference method.(2) In extracting the model of the moving target, according to the problem ofdiscontinuous outline of the moving target model during the extraction, firstly, a method ofmulti-objective separation of moving target was applied. Then start point was selected tocounterclockwise (or clockwise) close target outlines which could extract the moving targetfrom the original frames.(3) In tracking targets, when large areas of similar color interference existed in thebackground or the target was seriously kept out, CamShift tacking algorithm could notaccurately track the target, even led to the tracking failure. According to the problem existedin CamShift algorithm, a combinative method of CamShift and Kalman algorithm was provided in this thesis. The method used the Kalman filter to predict the searchingwindow-location of CamShift algorithm and in the mean time used the CamShift algorithmto calculate the optimal location and size of moving targets in the frame. When large areasof similar color interference existed in the background or the target was seriously kept out,the Kalman filter was used to calculate the target position instead of the predictive value inthe CamShift algorithm.(4) A fitting method was provided to attach the start point and the end point of thetrajectory points into a single line, and calculate the distance between the other points andthe line. If the distance was less than the given threshold, the fitting was successful.Otherwise connect this point with the start and end ones, and then do the same to eachparagraph respectively as above. Finally the fitting line of the track points was obtained.With the differential angle of the fitted trajectory instead of the curvature to describe thechange of the trajectory, the situation of the moving target was got through the analysis andcomparison. The experiment shows that this method can effectively detect target behavior.(5) A method of judging the motion of moving target was provided in this thesis.Analyzing the direction (reverse or stop) of the moving target to central points of a set ofcontinuous images and evaluating the speed of vehicles with two detection lines, which wasto judge that if the moving target was over speed.(6) In the combination of the research results, a system of detection and trackingtechniques for moving vehicles in video sequences and trajectory analysis has beendesigned.