Video Object Tracking System Based On Particle Filter Algorithm

Video object tracking is a hot and difficult problem in the field of computer vision, digital image processing, and machine learning. Thus, it attracts a growing number of researchers. Moving object tracking has been widely used in many fields, such as military, public security, video surveillance, intelligent transportation, medical diagnostics, and human-computer interaction. With the development of computer technology, the range of computer vision based application is growing and they will make important impact on the life of human beings.In the object tracking area, many practical applications contain various challenges including complex dynamic background, target occlusion, light intensity changes and so on. As a result, moving object tracking efficiently and robustly in complex environments is a difficult issue in computer vision. Particle filtering is a successful approach to deal with moving object tracking problem. However, like most computer vision algorithms, it has strengths as well as weaknesses. In this paper, we propose a new method to integrate advantages of the improved LBP and Mean Shift algorithm for high performance object tracking.Compare to the original particle filter, the addition of LBP model can make the tracking performance better because LBP embedded particle filter is not easy to lose the target when the target has been occluded in a short time. On the other hand, combining with LBP model makes the tracking process complex and time-consuming. Therefore, the Mean Shift algorithm is adopted to solve this problem. The Mean Shift method clusters particles to area of real state after re-sample. It not only makes particles closer to the true location, but also reduces the number of particles. Moreover, reducing particle number improves the efficiency of the proposed method and helps it runs in real-time. Experimental results demonstrate that the new method (MSLBPPF) outperforms the Mean Shift and the conventional particle filter in time-consuming and tracking accuracy. In addition, it remains high performance even in the case of short period target occlusion.