Moving Target Detection And Tracking Based On Binocular Stereo Vision

Binocular Stereo Vision is an important branch of computer vision. One important task of binocular stereo vision is to get 3-D information on objects under various conditions. Binocular stereo vision has unparalleled advantages compared to monocular vision. With the fast development of computer and electronics technology, real-time binocular stereo vision becomes realistic and it can almost satisfy the demand of real-world applications.Moving target detection and tracking is one of the most important subjects in computer vision, and it has been broadly applied in military visual missile guidance, video surveillance, medical image analysis, intelligent transportation and other fields. Since the traditional method is based on monocular vision, it has difficulties in tracking multiple moving targets, especially when the objects are occluded by each other. We adopt the binocular stereo vision method in target detection and tracking, and achieve results which monocular vision cannot provide. This thesis is primarily focuses on algorithms of detecting and tracking moving targets on a stationary background. The major research areas are summarized as follows:(1) Through research on stereo matching algorithms, analyzing advantages and disadvantages of the traditional matching algorithms, we propose a new stereo matching algorithm based on control-points constraints and area-correlation. First, corners are detected by the Harris corner-detecting algorithm. Second, stereo matching is used to obtain precise matching points, which are the control points. Finally, the non-corner pixels are matched by an area-correlation process method and under the constraint of control points. In this way, a dense disparity map is obtained. The method not only reduces the search matching spaces and improves the matching speed but also ensures the validity of matching.(2) Through research on the frame difference method and the background difference method, and analysis the advantages and disadvantages of traditional methods of target detection, we propose a novel detection method based on binocular stereo vision. It detects the foreground moving target by using the integration of disparity-based analysis and intensity-based analysis. This algorithm has the following advantages: it can detect targets correctly unaffected by changing lighting conditions. Shadows do not cause problems in the proposed algorithm. It can detect targets correctly when the objects are occluded by each other.(3) By analysis of the advantages and disadvantages of the traditional mean-shift tracking algorithm, we propose a novel method, which combines binocular stereo vision with an improved mean-shift tracking algorithm. By detecting target and removing the interference of the background, we track the moving target. We integrate the Kalman filter algorithm and traditional mean-shift tracking algorithm, and adopt different scale factors to combine the Kalman filter prediction results with mean-shift tracking results. The improved algorithm makes good use of space position of the target, so tracking reliability is enhanced. The method can track targets correctly unaffected by changing lighting conditions, shadows and target occlusion.