Visual Target Tracking Algorithm Based On Particle Filter Is Studied

As a major issue in Computer Vision filed, Visual tracking has its broad application prospect in military, monitoring, product control, bio-technology, etc. With soaring demands, visual tracking technology, especially its core algorithms, has been developed rapidly. In various algorithms, particle filter has received special attention because of its unique advantages in dealing with nonlinear problems. Along with rapid descending of computation cost, particle filter algorithm gradually becomes the mainstream. In this dissertation, particle filter is analyzed thoroughly and comprehensively, including some optimization and adjustment to apply it into real application. The main contents include:Firstly, to overcome the shortcoming of single visual cue in complex environments. A tracking algorithm based on adaptive cue fusion mechanism is proposed. The color cue, texture cue and edge cue are utilized to represent the target and adaptive integration strategy is applied to fusion these cues. Makes the tracking algorithm can adaptively adjust the weight of information based on the current track situation, to achieve the complementary strengths of the information. During designing particle filter based tracking algorithm, the likelihood model is constructed dependent on adaptive cue fusion mechanism, thus enhancing the robustness of tracking algorithm. The tracking results demonstrate that the tracking algorithm based on adaptive cue fusion is able to successfully track target in presence of move, rotation and partial occlusion.Secondly, the particle filter method has excellent tracking performance, but the algorithm also has some shortcomings, the most typical drawback are the large amount of computation and degradation of the particles, as well as particle diversity weakened because of using the traditional resampling. In this paper, to overcome the problems of the particle filter, the niche genetic particle filter algorithm has been proposed, in this algorithm, the Particle Filter and Mean Shift track the target at the same time, particles are sampled near the Mean Shift tracking results and find the weights, then with the merger of particles by the particle filter algorithm, the niche genetic manipulation has been adopted in the resampling part in order to increase the diversity of particles, to avoid particle degradation. Advantages of Mean Shift reduce the processing time, and the particles sampled by Mean Shift are closer to the true state because of taking into account the current observations. The experimental results show that the niche genetic particle filter algorithm which requires less particles can achieve a precise estimate of the state.