| Target tracking is a research hotspot of computer vision, and it has important practical value and broad prospects for development in both military and civilian fields. Particle Filter (PF) was introduced into target tracking because of the advantage in dealing with the nonlinear and non-Gaussian state, which also conformed to the actual situation of tracking environment. However, it will make the loss of the particles diversity, which leads to samples impoverishment and large computation problems when using resampling to solve the degradation phenomena in particle filter.Aiming at the problems above, the fundamental theories and the defects of the particle filtering are analyzed in this paper. An improved particle filtering algorithm based on the particle swarm optimization (PSO) was proposed and applied into the target tracking.1. Through comparing the characteristics between PF and PSO, the similarity in principle and the feasibility of the fusion algorithm are analyzed. Then, the theoretical basis of the improved algorithm is shown.2. The particle swarm optimization theory is introduced to particle filter, and a robust PSO-PF method is proposed. The state transition model is chosen as the simple first order autoregressive model, and the target is described by the gray level distribution of target area which is established based on the estimation of kernel probability density. By calculating the Bhattacharyya distance between the gray level distribution of the object reference and object sample, the observation probability model is constructed.It use the good local and global optimization ability of PSO to make particles in re-sampling particle sets co-operate with each other intelligently and alleviate the sample depletion. The experimental result shows that the presented algorithm improves the accuracy of target state estimation and reduces the computing time, and also has a better practicability with higher real-time performance. |
PDF Full Text Request