Research On Improved Target Tracking Algorithm Based On Compressive Particle Filtering

The technology of target tracking has become an important method in fields such as the aerospace, intelligent monitoring, safety management, medical and health. With the application being more and more extensive, the requirements of performance of target tracking technology is getting higher and higher. But in the process of target tracking, the problems such as object occlusion, scale and illumination changes are the important factors to influence the tracking performance. Therefore, in order to solve the problem of tracking robustness decline caused by occlusion and scale changing. An improved compressive particle filtering algorithm based on blocks matching and an improved algorithm fuse the window update mechanism has been proposed in this thesis to solve the above two questions.The target tracking algorithm based on particle filtering can tracking target with a high accurate when the target was occlude, because the particle filtering tracking algorithm can obtain the posterior density function based on the Monte Carlo simulation. If we want high accurately, we must increase the number of particles, but the calculating complexity will be increasing with it. So the scene of apply will be restricted. With the maturity of compressive sensing theory, we can reduce the dimensions of target picture with the compressive sensing theory, so the efficiency of operation will be decreasing in a large level.Considering the fact that the target occlusion will appear in the practical application, the two algorithms mentioned above can achieve complement on theoretical. So an improved compressive particle filtering algorithm based on block matching has been proposed in this thesis. The improved compressive particle filtering algorithm is based on the framework of particle filter. In view of the complexity of blocking, block matching mechanism has been proposed, Determine the parts which were occluded according to the similarity of sub-part information. When the target was occluded, updating the target templates which sub-part were not occluded. In addition, the improved color feature extraction method fuse spatial information is used to describe the target feature extraction. Through the improvement of the above, it alleviates the problem of tracking drifting and even tracking loss due to occlusion in some degree.In the process of tracking, target block is often accompanied with the change of target size. An improved algorithm with the window update mechanism was proposed. In this new algorithm, we fuse the window update mechanism in the target tracking algorithm which was proposed in the section three. We set multiple tracking window, and locate the target position with the highest matching data between the target template and candidate template after adjust the size of windows in present frame. This algorithm can realize the target tracking not only the target size was changed but also occlusion because it based on improved compressive particle filtering algorithm.The qualitative and quantitative analysis of the two improved algorithms are presented in this thesis through theoretical and simulation experiments. We can make conclusions that the improved compressive particle filtering algorithm based on block matching can reduce the influence of occlusion. In addition, the improved algorithm with the window update mechanism can resolve the problem of occlusion and the change of target size at the same time. Experiments indicated that the improved algorithms can guarantee the efficiency and accuracy.