Improved Sampling Importance Resampling Particle Filter Optimized by Tabu Search

Abstract

How to get better performances with fewer particles and shorter calculation time is a difficult problem of Particle Filter (PF). Some good choice of importance density and resampling are proposed to resolve the degeneracy phenomenon, and also, can reduce the particles number. Sampling Importance Resampling Particle Filter (SIR-PF) is a version of PF. In this article, TS (Tabu Search) is only used as part of initial particle choice step of the SIRPF (Sampling Importance Resampling Particle Filter), and not used in resample step, so the calculation time of the main program would not be added. The initial particles of SIRPF are chosen by tradition sampling method and TS. The improved SIRPF is tested with a simulation of a land vehicle navigation system, and compared with the tradition SIRPF, with the same amount of the initial particles. The calibrating performances of the algorithms are: Location precision, divergent times and calculation time. The results show TS can improve the particle quality, and the SIRPF performances are improved. The improved SIRPF with 300 particles has the similarity better performances as the tradition SIRPF with 1000 initial particles and need less calculation time.