Algorithm Research About Detection And Tracking Of Vehicle In Surveillance System

Now ITS(Intelligent Transport System) is an important research direction and hotspot of computer vision. With the gradually development of the computer vision, ITS is growing up in these years. The detection and tracking of video vehicle are the core and foundation parts of ITS. They make it possible for obtaining the traffic parameter. Now the technology of the detection and tracking of moving vehicles is not mature, so we need to do research and improve the performance. In this paper, we have did some constructed work in these aspects.In this paper, we propose road-based detection of moving vehicles. When detecting the moving vehicles, because of the influence of the changing illumination, vile weather and the vibration of trees, we can not detect the vehicles accurately. By investigating the distinction between the motion vectors of the dynamic background and those of moving vehicles, we find the fact that the motion vectors of the dynamic background disperse in a large region because of the uncertainty of the motion while those of moving vehicles cluster in a small region because of the certainty of the motion. Based on this characteristic, Gaussian motion model is proposed to model the motion of the moving vehicles in different roads and that of the dynamic background. First we get the motion pixels by the background subtraction. Then we classify them with Bayesian framework.In this paper, we propose the moving vehicles tracking based on the transform model between the 3D actual scene and 2D image coordinate. When obtaining the characteristic, we get the position of the centroid of the moving vehicles, the width and height of the tracking windows, the area of the moving vehicles and average intensity of the moving vehicles. While in the forecast of the position and shape of the moving vehicles, we propose the quadratic of the motion locus based on the imaging characteristic of the camera, which can forecast the position more accurately. We make the matching vehicles well between consecutive frames integrating many characteristics. When updating the parameters of model, we propose the concept of three zones. In every zone, we process the moving vehicles in different ways. At the same time, we propose some methods in initializing and updating parameters.