Moving Objects Detection And Tracking Based On Traffic Video

With the development of the traffic, the process of the traffic video becomes more and more important. We lay an emphasis on the research of moving target detection and tracking in traffic video with stationary background. At last we discuss the feature distilling and classifying in the view of the system.Target detecting in videos involves verifying the presence of an object in image sequences and possibly locating it precisely. Target tracking is to establish the target relation between image sequences and verify movement of target such as appearance, disappearance, occlusions and speed of target. Classifying and recognition is distilling feature of tracking target then getting result of classifying. Target detecting, tracking and recognition is a closely related process, result got from detecting influencing tracking and recognition directly. Due to the difference of real environment, changes of the illumination and shadows there are great difficulty in target detecting, tracking and recognition.We construct a moving target detection module, which is suitable for traffic video, using GMM based on the summarizing usual target detecting methods. We present a method which is used for initialize the background. The whole process of target detection has four step: initializing background, moving target detecting, morphologic filter and signing connected field. The testing result proves this method is useful.We tried Kalman filter, Meanshift for tracking purpose. We analyzed their academic basis, realized these two methods, and unified the interface with detecting module and classifying module. We present a tracking method, which is suitable for traffic video, based on the rules. We construct rule depend on the difference of human and vehicle in speed and shape and the similar of the historgram of the target template and candidate target. This method is simple, effective and has high speed. It is suitable for traffic system.Classifying used ratio of area of adjacency rectangle and area of target, ratio of the area and girth, ratio of the height and width of the rectangle as feature. We tried template matching and RBF ANN as classify method. At last we realized a surveillance system using detection module, tracking module and recognition module.