Hybrid Particle Filter Resampling method based on Particle Swarm Optimization and Genetic Algorithm for Track Before Detection

Abstract

The estimation of state parameters for nonlinear and non-Gaussian systems the sequential Bayesian filtering approach used is the particle filtering. Particle filter cannot determine these parameters directly; it can relate the noisy observations with the probability masses. The particles are generated with the previous knowledge, with their weights evaluated from the likelihood of being the true value of those parameters. Most of the particles have insignificant weights. Every particle filter will fallow these corresponding steps generation of particles, calculation of their weights and regeneration of particles called resampling. During the nonlinear state estimation, the resampling step the degeneracy problem is solved some extent. During the resampling process because of replacement of low weight particles with high weight particles sample impoverishment problem occurs. By analyzing the causes of sample impoverishment, particle swarm optimization (PSO) algorithm is introduced to particle filter, this is a swarm intelligence algorithm. During the resampling step it enrich the diversity of sample set. But the variance of measurement noise significantly affects the fitness function of PSO. This will lead to limits the accuracy. In this paper the PSO is combined with the genetic algorithm (GA). This hybrid algorithm combines PSO fast convergence speed with the robust global searching ability of GA to increase the diversity of the particles while safeguarding the effectiveness of superior particles, and improves the accuracy of finding the finest solution.