Object Tracking Using Genetic Evolution Based Kernel Particle Filter

摘要

A new particle filter, which combines genetic evolution and kernel density estimation, is proposed for moving object tracking. Particle filter (PF) solves non-linear and non-Gaussian state estimation problems in Monte Carlo simulation using importance sampling. Kernel particle filter (KPF) improves the performance of PF by using density estimation of broader kernel. However, it has the problem which is similar to the impoverishment phenomenon of PF. To deal with this problem, genetic evolution is introduced to form new filter. Genetic operators can ameliorate the diversity of particles. At the same time, genetic iteration drives particles toward their close local maximum of the posterior probability. Simulation results show the performance of the proposed approach is superior to that of PF and KPF.