Strategy Of Vehicle Tracking On Urban Tunnel Monitoring System

With the development of the city,the emergence of the tunnel eased the traffic pressure to a certain extent.In order to ensure the normal running of the vehicle in the tunnel,monitor the abnormal situation and be able to timely treatment,the need to establish a set of more perfect tunnel video surveillance system.In the video surveillance system,the detection and tracking of the vehicle is the key technology.This thesis mainly studies the vehicle detection and tracking algorithm in the surveillance vedio of the urban tunnel.The combination of off-line training and on-line learning to improve the accuracy of the vehicle tracking,so as to better monitor the operation of the vehicle in the tunnel.First of all,the analysis of the actual situation of city tunnel in video surveillance,this paper puts forward a suitable environment in tunnel denoising method can preserve the edge details of the image features better in denoising and.Then through the analysis and comparison of several vehicle detection algorithm is applied,the adaptive background model,improved background update strategy,namely the use of the background of the previous frame to the current frame of the foreground and background are separated,and the previous frame and the current frame corresponding to foreground pixels do statistical average,to construct the background image,so that the result of detection in the maximum extent to ensure the authenticity and real-time of image background,the morphology to improve the extraction area of the vehicle.In the off-line training phase of the vehicle classifier,a multi feature fusion method is proposed,which is based on the analysis of the characteristics of the vehicle in the tunnel environment.The method for extracting the color features of the vehicle,edge feature and Haar-like features to train weak classifiers,makes up the shortage of single feature,to the classification error rate to give the corresponding threshold,so as to accelerate the training speed of a classifier offline.In order to better test the classifier’s classification and detection ability,by way of a serial concatenation will eventually cascade classifier in the target area of the vehicle judgment,compared to single feature classifier,the classification accuracy increased about 10 percent.In the vehicle tracking stage,in view of the existing tracking methods can not adapt to the environment of tunnel traffic vehicle scale changes and the tracking problem,put forward a based on the combination of boosting the off-line and on-line learning of moving vehicle tracking method.The method first used offline multi feature classifier to judge to extract the sense of whether there is a vehicle in the region of interest;then in the online tracking of the process to the offline classifier as the initial selector and the meanshift method estimates the initial position for reducing the search time and regional and through offline classifier to deal with the target vehicle with lost.The experiment results show that,compared to the error tracking algorithm in online-boosting algorithm is reduced by 25%,decreased by 5% than the TLD algorithm.In the multi vehicle tracking can also be well adapted to the tunnel environment.