Intelligent Video Surveillance, Moving Target Detection And Tracking Technology

Intelligent video surveillance (IVS) is a multi-disciplinary research involving the forefront in area of image processing and computer vision .IVS has important scientific significance and broad prospects, at the same time it is full of enormous challenges. Moving target detection and tracking is the most basic core technologies of IVS, which not only are the cornerstone of the following advanced processing and application, such as target classification ,behavior analysis, event detection, behavior recognition, the video image compression and semantic index and so on, but also are the key for automation, intelligent and real-time application of a video monitoring system . How to find an algorithm which can detect and track targets accurately, rapidly and stable, which also can deal with the complex environmental changes, is still a big issue to be resolved.In this paper, based on the background of achieving intelligent video surveillance system, the author focus on the research of moving target detection and tracking, which are key technologies for IVS. The main work of this paper is following:The author summarized the moving target detection algorithms, and had an in-depth analysis on the subtract background method based on mixed Gaussian model, and described the basic theory and the simplification for concrete application .The experimental showed the efficiency of the method in dealing with multi-modal background. Next the author outlined the algorithm for shadow detection, and had a detailed analysis of the shadow detection algorithm based on HSV color space.Then the author had an in-depth analysis of the particle filter, and described the basic algorithm framework ,under which the particle filter could be used for tracking .More cues on the integration of the tracking method was analyzed. In the framework of the particle filter, a tracking method based on adaptive integration of color cue and corner cue was proposed. Experimental results showed that the proposed method is more stable than the methods which only using a single cue, and can effectively handle the problem of suddenly change of light and the background interfusion.The author had an analysis on the occlusion problem of tracking, outlined the state-of-the-art of the methods to solve the occlusion problem. A novel adaptive multiple-cue fusion based algorithm was proposed for anti-occlusions object tracking, which was in the framework of particle filter. Binary shape template and color histogram based on kernel ellipse were used as cues for tracking, the visibility of an object was introduced to evaluate the reliability of the color information under occlusions, according to the visibility the two cues were fused adaptively to obtain the measurement model. Experiment results showed that the proposed algorithm was much more robust to occlusions problem, compared with the tracking method based on single cue.