Design And Implementation Of Moving Object Detection And Tracking System Based On DSP

The moving object detection and tracking technology is one of the key technologies in the intelligent video surveillance application, by which the surveillance system can detect active changes in the monitoring scene, compute motion traces of moving objects, and provide the necessary information for the high level motion analysis. Due to the strong demand for the improvement of the intelligence degree of surveillance equipments and the increasement of the social security level, the moving object detection and tracking technology will have a bright application prospect.This paper is focused on the design and implementation of the moving object detection and tracking system based on DSP, including the design and implementation of the embedded video process system based on DSP DM642, as well as the research and implementation of the moving object detection and tracking algorithm. The main work and contribution of the thesis are as follows:Firstly, an embedded video system based on DM642 is designed and implemented. The SCH and PCB of peripheral hardware modules, the memory module and the video processing module are given and analyzed respectively. Further, the major problems and debug experiences are summarized.Secondly, the transformation process and main features of video signals in each phase are analyzed. Based on the discussions of the basic procedure and work mechanism of the real-time video process, the driver software is developed. At last, a conclusion is drawn for the DSP software debugging method.Thirdly, the whole process of the detection and tracking system is analyzed, the algorithms of the different stages are researched and implemented, and the corresponding experiment results are given. For the single modal background, a novel background subtraction technology by using the frame difference is proposed. For the multi-modal background, a detection algorithm by utilizing the multi-subspace of histogram is proposed. In the phase of the moving object tracking,this paper presents a hybrid approach based on KALMAN filter and histogram matching.Finally, the work is briefly summarized and some remarks on further research are also presented.