Study On Video-based Vehicle Detection And Tracking Techniques In Urban Road

With the development of city and economy, the burden of the road transportation system becomes more and more heavy. Therefore, more attention has been paid to the vision-based intelligent transportation system. The key problem of this application is vehicle detection and tracking from the video sequence captured by static traffic surveillance camera. Moreover, there are many problems, such as shadow of moving objects, change of light, vehicle sheltering each other, and vehicle occlusion, which make many difficulties to the vehicle video detection and tracking system. With those problems, in this paper, we do some researches on the key technologies of vehicle detection and tracking, which will play the important role in the vision-based intelligent transportation system. Our research works are summarized as follows.Firstly, we make a comparison among the common methods of moving vehicle detection. According to complex urban road traffic scenes, a new method for moving vehicles detection is given based on background extraction of Gaussian mixture model, which can attain and update the background adaptively by Gaussian mixture model, and integrate background subtraction and improved symmetric interframe subtraction to detect moving vehicles. The problems of initializing background difficulty and updating background slowly are resolved in this method, and it can be adapted to sudden change of light. Moreover, this method increases the accuracy of vehicle detection. The experimental results show that the method can detect vehicle under the scenes of urban roads effectively.And then, according to the problems of moving shadows, a criterion which is used to decide whether running the algorithm of shadow elimination or not at this moment is given: shadow is detected by evaluating the ratio between the area of vehicle and the area in the vehicle's convex hull, and the algorithm of shadow elimination is run or not by evaluating the shadow regions on statistics. This criterion of the method improves the whole system real-time. Additionally, a new algorithm of shadow elimination is proposed which combines the texture and chrominance properties of the moving foreground regions. In this algorithm, texture-map, luminance-map and chrominance-map of each moving foreground region are extracted respectively. An OR-map is then constructed by performing a logical OR operation of the three maps. Finally, the shadows of moving vehicles are eliminated effectively through running a series of morphological operations on the OR-map.In the end, according to the problems of complex vehicle mobility, vehicle occlusion and disappear in seconds under complex urban road traffic scenes, a regional model matching vehicle tracking method is applied, which is based on Kalman prediction model. This method can predict motion state of vehicle and remember tracking effectively. Comparied with the common methods of moving vehicle tracking, this method has advantages of reducing the seeking region of vehicle matching, predicting the position of vehicle optimally, and resolving the problem of vehicle complete occlusion. In this method, a regional model of each moving vehicle is established by extracting centroid, width and height parameters of the moving foreground region, and updated by Kalman filter. The moving vehicles location and tracking are completed accurately by two matching criterions. Combination of this vehicle tracking method and reasoning model, the problem of vehicle complete occlusion can be resolved. According to the problem of vehicle partial occlusion, the detected partial occlusion is eliminated by finding a "parting line" in each partial occlusion foreground region.