Research On Vehicle Detection And Tracking Technology In Video Intelligent Monitoring

In recent years, with the increasing amount of vehicles in our country, effective regulation of the vehicle has gradually become an important problem to be solved urgently for the related management. For the needs of the development of security industry, video surveillance has developed rapidly in recent years, and using video surveillance technology to monitor vehicles in real time has become an important form of vehicle monitoring. To make up for the limitation of the traditional artificial video surveillance, integrating the computer vision technology into the video surveillance system has become a hot research topic in video surveillance. In the intelligent video surveillance system, the detection and tracking of vehicle is the basis of vehicle behavior analysis and understanding,and only in the good detection and tracking of vehicle can further expand other applications in video monitoring system. So this paper mainly focuses on the vehicle detection and tracking in video intelligent monitoring system, and has finished the following work:(1) In the vehicle detection section, the vehicles in video surveillance are detected by the vehicle cascade classifier which is trained by the Adaboost algorithm, and the feature used to train is the Haar-like feature of sample image. The algorithm firstly collects the positive and negative samples of the vehicle to set up sample set, and then computes the feature values of the sample set fastly by integral graph, finally after setting the training parameters, the algorithm begins to train the vehicle cascade classifier. The detection result on the universal vehicle data set shows that the vehicle cascade classifier trained in this paper has a good performance.(2) In the vehicle tracking section, a vehicle tracking method based on Kalman filter is us-ed to realize multi-vehicle tracking in video surveillance system. The method firstly locates the center of vehicle rectangular area which is detected by the classifier in current frame and then sets the Kalman estimation model to predict the center coordinate of vehicle rectangular area detected in the next frame, the state parameters input into the Kalman estimation model are the coordinate and velocity of the center in current frame. By establishing matching criteria, the predicted center coordinate of vehicle rectangular area and the practically detected center coordinate in the next frame are matched to decide whether it is the same vehicle. At last, the method updates the Kalman estimation model based on the matching result and conti-ue to predict, thereby the method sets the tracking chain to completes the tracking of the multi-vehicle under the video surveillance.(3) At last, this paper used the surveillance video captured from monitoring scene to test t-he vehicle cascade classifier and tracking algorithm. The test result shows that the vehicle cascade classifier has a good detection rate for the most vehicles that meet the position requirements of the vehicle in the surveillance video, and tracking algorithm is also able to track the vehicles detected stably.