Distributed genetic resampling particle filter

Abstract

Particle filter (PF) is one of the most important nonlinear filtering methods and has received much attention from many fields over the past decade, but the degeneracy phenomenon and large computation amount of PF have significant negative impacts on its filtering accuracy and real-time performance. In order to solve the problems of PF, this paper integrates distributed genetic algorithms (DGAs) and PF, and puts forward the distributed genetic resampling particle filter (DGRPF). This method divides all particles into several subpopulations to parallel execute particle filtering. Several genetic operators such as crossover, mutation, selection and migration are adopted to optimize the resampling process, which can effectively suppress degeneracy phenomenon, increase particles diversity, and make PF easy to execute in the distributed processor. By software simulation, DGRPF is compared with several existed PF algorithms in the tracking performance, estimation accuracy and computation efficiency, and the effectiveness of DGRPF has been verified.