粒子滤波器重采样算法的分析与比较

Analysis and Comparison of Resampling Algorithms in Particle Filter

重采样算法解决了粒子滤波器的退化问题。从重采样实现原理、均匀分布理论和计算复杂度的角度对目前广泛应用的四种重采样算法(包括多项式重采样、残差重采样、分层重采样和系统重采样)进行了理论分析,同时引入有效样本数,设定重采样阈值,将四种算法应用于经典纯方位跟踪,通过仿真比较不同重采样算法条件下的运行时间和跟踪性能,并分析重采样阈值的设定对滤波性能的影响。仿真表明,四种算法在跟踪性能上很接近,系统重采样和分层重采样算法下的运行时间很接近,优于其余两种算法,设定重采样阈值后,整体跟踪误差减小了约50%,但并未明显减慢跟踪误差随时间推移的发散趋势。

Resampling algorithm solves the degeneracy problem of the particle filter. In the points of realization theory, resampling quality and the computation complexity, the four widely used resampling algorithms （including multinomial resampling, residual resampling, stratified resampling and systematic resampling） were theoretically analyzed. And meanwhile, effective sample number was introduced, resampling threshold was set, these algorithms were applied in classical bearing-only tracking, by simulation tracking performances and running time of the particle filter were compared, and the effects of setting resampling threshold on filtering performance was also analyzed. Simulation shows that tracking performances of these algorithms are similar, while as to the running time, stratified resampling and systematic resampling are approximately the same, obviously superior to the other two algorithms, and setting resampling threshold could reduce wholesome trackin~ errors bv about 50%. but couldn＇t obviouslv delay tra^kin~ errnr div^r~n~ ~n^rl nl,~n,~ tlm,~