Research Of Object Tracking Based On MeanShift Algorithm

Moving target tracking is a highly cross-disciplinary, which involves many fields, such as computer image processing, video image processing, pattern recognition, artificial intelligence and so on. The diversity of the target characteristics, complex background and the occlusion problem in the process of tracking are key factors to restrict the robust of tracking algorithm. Therefore, the research of real-time and robustness is still a great challenge in the field of object tracking.In this thesis, The target tracking problems which based on the original MeanShift algorithms are studied firstly, and the experimental results verify that the original MeanShift algorithm could make the target tracking failure under the complex background and occlusion circumstances. On this basis, An improved tracking algorithm is proposed. That is, the candidate target template's center location of tracking target area is calculated by the centroid of each color. Compared with original Meanshift algorithm, the improved MeanShift algorithm includes the spatial information on color distribution. According to the improved algorithm, The problem of original MeanShift algorithm which uses color histogram as the target feature and does not take the objectives of spatial information into account is solved.Secondly, Research on the occlusion circumstances during the tracking process. Instead of using single fixed target model, a selective sub-model update strategy is analyzed and discussed. In order to increase the robustness of tracking process, The selective sub-model update strategy is added to the improved MeanShift algorithm. This new method solves the failed tracking problem under the occlusion circumstances. Furthermore, experimental results show that the improved tracking algorithm is better than the original MeanShift algorithm, Especially under the condition of complex background and occlusion circumstances.