Research On Particle Filter And Correlative Algorithm In Object Tracking Of Sport Video

Object detecting and visual tracking are an important direction in computer vision field. It's a great challenge to investigate a robust, precise, stable, and applicable object tracking algorithm, which is both theoretically and practically valuable. Nowadays, the programs of sports video are becoming more and more popular for their special fascination. Therefore, it possesses high actual meaning and commerce worthiness to analyze the sports video.Firstly, the thesis narrates the Bayesian filter theory and the particle filter theory, discusses the main questions of the particle filter and how to solve them. Each step of common particle filter is explained in detail. Experimental results have indicated that the particle filter algorithm is better than the Extend Kalman Filter algorithm, when they faced to Non-linear and Non-Gaussian systems.The thesis discusses CAMShift algorithm and theory in detail. With CAMShift algorithm targets object in video sequences, we have summarized the advantages and the disadvantages of this algorithm.The particle filter tracking method, based on color feature template, is designed, and object modal in it has the ability of automaticity and update. The theoretical equation of the relation of color histogram match degree,particle number and particle diffusion radius are proposed. With the results of tracking object experiment, the quantitative analysis has shown the effect the particle number and particle diffusion radius on the precision in object tracking, and summarized the experience equation of their relation. This study has a significant guidance in the video tracking field based on the particle filter.On the basis of above study on particle filter, the thesis proposes the self-adaptive particle filter algorithm. The algorithm adopts effective measure about object occlusion and tracking after occlusion and resumed tracking of the losing object. Then, to natural tracking object, the algorithm reduces particle number necessarily, which offers an important clue to real time tracking object.The experiments of tracking sports video of two styles indicate that proposed adaptive particle filter algorithm in this thesis will satisfy the quasi- real time and robust of tracking object on a certain extent, which pioneers a new method for tracking object of sport video.