Dsp-based Movingtarget Recognition And Tracking

The video's moving target recognition and tracking is an important research direction in the field of computer vision. It is the key technology of the security monitoring, the weapons guidance, and the automatic production line's control system. The research's key point is the moving target recognition and tracking system's accuracy, robustness and real-time. This dissertation studies on how to improve the accuracy of the system and how to improve the speed of the system. This dissertation mainly includes the following content:(1) Research of the moving target recognition algorithms. The moving target recognition means that the moving target is extracted from the video, and the step is a basis step for subsequent processing. Because the real world is very complex, it is very difficult to accurately extract the target from the video. This dissertation proposes a new target recognition algorithm based on the multi-Gaussian mixture model and the YUV color space. The algorithm can effectively eliminate the impact of the global illumination change and the impact of the background noise. Experiments show that the algorithm not only extracts the moving target from video effectively, but also overcomes the disadvantage of multi - Gaussian model, which is sensitive to global change.(2) Research of the moving target tracking algorithms. This part mainly researches on these typical moving target tracking algorithms, and compares their respective advantages and disadvantages. Considering with the real-time requirement of the moving target recognition and tracking system, the KLT algorithm based on feature tracking is selected to complete the moving target tracking. Finally the system meets the real-time requirement.(3)Due to the complex calculation of the moving target recognition and tracking, we use a DSP processor to implement these algorithms. We complete the experiment with the TMS320DM642 chip which is dedicated to image processing. We also give the strategies including parameter setting, algorithm implementing and optimizing on the board.Through deep researching the moving target recognition and tacking algorithms and the algorithm's realization method on DSP, This dissertation achieved a high accuracy and real-time moving target recognition and tacking system.