Research Of Particle Filtering Algorithm In Video Target Tracking

Video target tracking is a very important work in computer vision, whichattracted more and more attention from researchers. It has widely applied in thefield of intelligent visual surveillance, human-computer interaction, automaticdriving, military security and so on.The particle filtering has excellent robustness with various forms as well as inthe nonlinear motion and measurement model, which can solve the problem ofnon-Gauss’s state estimation in video target tracking. But there have problems ofparticle degradation and scarcity caused by the re-sampling which affects thetracking accuracy, especially in the case of occlusions and fast moving. In view ofthese questions, this paper mainly focuses on improving the particle filteralgorithm in order to improve the performance in real-time, robustness andaccuracy of target tracking method.To improve the accuracy and robustness of occlusions and fast moving invideo target tracking, a tracking algorithm based on particle filter optimized by anew cloud adaptive particle swarm optimization（CAPSO） is proposed. Thepossible position of moving target in the next frame image is predicted by particlefilter, and matching the target template and candidate regions with the colorhistogram statistical characteristics to ensure the tracking accuracy. Then theproposed CAPSO is utilized to divide the particles into three group based on thefitness of the particle in order to adopt different inertia weight generating strategy.The inertia weight in general group is adaptively varied depending onX-conditional cloud generator. The inertia weight has randomness propertybecause of the cloud model. Therefore, the re-sampling frequency of particles filteris reduced, so that the computational cost of particle filter is effectively reducedand it is effective to solve the target tracking problem of occlusions. In addition,the algorithm can effectively balance the global and local searching ability of thealgorithm by adopting three different inertia weight generating strategy, which can adjust the particle search range, then can adapt to different motion levels.A Quasi-Monte Carlo particle filter algorithm based on good point set(GPS-QMCPF) is proposed for overcome the drawbacks of large computationalcost and poor real-time performance in video target tracking. In the proposedalgorithm, a new Quasi-Monte Carlo sequence is constructed by using the principleof good point set in number theory. Considering that the good point set is morehomogeneous distribution and lower discrepancy than the standard QMC sequenceand the random sequence, GPS-QMCPF can obtain a faster convergence speed inthe filtering process and a better accuracy of the state estimation. Furthermore, there-sampling frequency is reduced, which results in a lower computational cost. Theproposed algorithm gets a more accuracy estimation than standard QMC filter andparticle filter in system state estimation, and in video target tracking application,the proposed algorithm possesses advantages of good tracking accuracy andreal-time standard, even in case of occlusions.