Study On The Critical Technologies Of Particle Filter And Its Application

Particle filter is a reasoning algorithm based on Monte Carlo methods and Bayesian. It can be applied to any nonlinear non-Gaussian systems which can be represented by state space model. Particle filter is flexible and easy to program, so it attracts widely attention, it has become research focus of other fields such as the signal processing, artificial intelligence, automatic control. The study of particle filter is still in its infancy, many of the key technology such as proposal distribution, re-sampling, convergence analysis do not have effective solutions. In this paper, in view of question of the degradation and impoverishment, from the basis of particle filtering theoretical we study the key technologies of particle filter such as resampling, adaptive mechanisms, diversity measure, and single target tracking applications.This article first conducts the deep research to the particle filtering theory and the algorithm, and introduces in detail the principle and the step of particle filter algorithm, then four kind of classical resampling algorithms - multinomial resampling, stratified resampling, systematic resampling and the residual resampling have been carried out on the theoretical analysis, and the diversity and performance of four resampling algorithm are compared through simulation. Then it introduces partial resampling algorithm, performance of the algorithms were compared from theoretical analysis and simulation when the threshold in the different weights.For the sample impoverishment which caused by resampling, based on partial resampling, a kind of improvement algorithm -weight choose restructuring resampling based on the weight optimum are given. After the groups step ,the weight particles which need resampling are made optimum composition, the originally small weight after weight optimum composition had been enhancement, it effectively relieve sampling impoverishment question. The feasibility of algorithms are verified under the simulation of single-target trackingFor how to effectively control the number of resampling to improve the robustness of filter in this paper, two adaptive time resampling algorithm are given combined with the adaptive and diversity measure mechanism, adaptive partial systematic resampling algorithm and based on diversity guidance adaptive resampling. The two algorithms basis on the degree of the particles weight deterioration to adaptive tuning resampling time which reduces the number of reasampling and relieve sampling impoverishment question. The effectiveness of algorithms is verified under the simulation of single-target tracking.Based on the study of rsampling, the adaptive mechanisms and diversity measure, two improved particle filtering algorithms based on adaptive mutation are propose, adaptive mutation particle filter based on diversity guidance and adaptive mutation particle filter based on weight choose restructuring resampling. Both in the two improved particle filtering algorithms, the particles after resampling add adaptive mutation step, which can effectively improve diversity and estimation performance of the filtering. The effectiveness of algorithms are verified under the simulation of single-target tracking.