Research Of Tracking Algorithms Based On Low Load DSP Platform

On basis of the DSP + FPGA architecture of digital image processing hardware platform with the high performance DSP chip TMS320DM642 as core, this paper finishes the research of object tracking algorithms and implements the algorithms in the low load DSP platform of image tracking system for MUAV.In the first part, the thesis mainly researches on image tracking system working principle and ideas of structure design, and then proposes the hardware block diagram of image processing unit. The video stream signal standard of BT.656 is analyzed from aspects of sampling frequency, resolution, transmission format, the video timing and reference code as well as the quantities of data. The system software process and method for control parameters of the PTZ is given on the basis of research of system working principle.The second part which is the major focus of the paper is on real-time tracking algorithm. For requirement of real-time, the paper first studies centroid tracking algorithm, pointing out the disadvantage of classical segmentation algorithm and proposing a iterative segmentation algorithm which achieves better effect of segmentation. But the centroid tracking based on image segmentation can only be applied in the context that relatively simple. Therefore, the paper develops fast correlation match using adaptive template tracking algorithm which combines improved mean removal correlation with the diamond search algorithm, adding weighted modification strategy and adaptive template update strategy to improve the algorithm in real time and stability.In recent years, particle filter theory shows a strong vitality and superiority in dealing with problems of nonlinear systems. After deeply understanding the theory of particle filter, we build a tracking framework based on SIR particle filter, using gray-scale distribution of the target as the target characterization, and propose a tracking algorithm based on SIR particle filter. Finally, we make simulation on the PC and transplant the algorithm to DSP system. The simulation experiments verify that the tracking algorithm based on particle filter is effective and stable, and fully meet the system requirement of real-time. Compared with the traditional correlation tracking algorithm relying on the"peak"information, the tracking algorithm based on particle filter demonstrates the advantage of "multi-peak" track, which greatly improving the robustness of tracking.