Research On Detection And Tracking Of Moving Target Based On Meanshift

Moving target detection and tracking is an important research direction in computer vision, and plays an important role in the military, aerospace, criminal investigation, sports and traffic management, has brought the huge research value and commercial value.In this paper, a comprehensive and in-depth on the detection and tracking of the content of the research and analysis, the focus is on learning and improvement of MeanShift algorithm.1. For the three kinds of moving targets detection algorithm and the classical MeanShift algorithm commonly, described the principle and calculation process in detail then summarized several limitations of the MeanShift algorithm and provides some solutions.2. According to the scale change in the traditional algorithm, the computation of reverse projection, also consider the weight distribution of pixel reset in the target area, method of pixel conversion from the center and the boundary of the important degree, in order to obtain the pixel scale change with new information, so that the selected track box using color histogram pixels, then the position of the center of mass frame tracking method used to update iteration, proposes a kind of adaptive window changes based on MeanShift algorithm. The final proof can achieve the adaptive scale change tracking effect in the experiment.3. According to the condition of occlusion is studied and improved, prepared using moving target block and prediction mechanism and a moving target block, through the increase in the target area in rectangle, and the method to obtain the sub region; and in the initial target center position and new position vector distance sub region; using the similarity ratio by weight, comprehensive weight and the new sub regional center weighted to calculate the new position finally; judging Bhattacharyya coefficient and the occlusion threshold to determine whether affected by occlusion again then proposes a kind of resistance occlusion moving target tracking based on MeanShift algorithm, experiments show that the new algorithm to deal with the effect of occlusion is good.Based on the traditional MeanShift algorithm in the paper, two problems of the scale change and occlusion are improved through the experiment to prove the validity of the new algorithm and stability, solved them basically.