Moving Target Detection And Tracking Key Issues,

Detecting and tracking of moving objects are two core processing techniques of intelligent video surveillance systems, which are the key to achieve a intelligent of surveillance systems. In this thesis, we focus on key problems of moving object detection and tracking in dynamic scene.The main includes removal the moving shadow,signal camera of target steady tracking and occlusion handling, non-overlapping multi-camera objectives associated match etc.Firstly, we do much studying by analyzing the dynamic shadow characteristics in image sequence. We refine the method of removing shadow in rgb color space by chromaticity, normalized brightness and mean changes of Local Change Average in pixel color value,through analyzing the removing shadow insufficiency based in attributes as chromaticity and LBP texture.Secondly, we do much studying in steady tracking, an improved in RGB color feature relevance matching method is presented. We extract normalized UV chromaticity and the edge gradient direction in eight histogram as basis features template for matching in tracking. In signal camera, we put forward method to solve object tracking by motion state prediction of object based on Kalman filter and features template matching. That is no occlusion in moving objects, improved the efficiency of tracking algorithm by searching match object around the prediction position of tracking object. That is sure of occlusion, an method is presented by increasing successive frames can be tolerated, in which keeping continuous tracking of targets by only prediction and correction the motion state of object while updating features template by threshold. On the basis of the successful object tracking in single camera, we put forward method to achieve objectives associated match in non-overlapping multi-camera by accumulating features template of moving object. This method can reduce influence of light and the posture changing on the target characteristics.Finally, we test our methods in real scene. Experimental results show that removing shadow algorithm proposed can keep the best foreground and remove shadow accurately. Furthermore, object tracking method proposed can achieve object tracking accuratly and timely in signal camera. Based on the successful and steady object tracking in single camera, the method of accumulating features template can keep continuous tracking of the same target in non-overlapping multi-camera. And the efficiency and robustness of my research results are verified by experiments.