Video Object Tracking Based on Swarm Optimized Particle Filter

Classical particle filter needs large numbers of samples to properly approximate the posterior density of the state evolution, and moreover, sample impoverishment is an inevitable problem, which is a key issue in the performance of a particle filter. In this paper, particle swarm optimization (PSO) was embedded into generic particle filter framework to achieve more robustness and flexibility. Samples were generated to represent the initial state of the object. Particle swarm optimized the sample set after prediction step. The object was tracked if the samples had reached convergence. Target state estimation was computed according to the globally best location of the entire population. Experiment results demonstrated that particle swarm algorithm can effectively eliminate particle degeneration and enhance robustness. Consequently the efficiency of video object tracking system was effectively improved.