Resampling in Particle Filtering - Comparison

Abstract

The article presents over 20 different types and variants of resampling methods. Pseudo-code has been added for a description of each method. Comparison of methods has been performed using simulations (1,000 repetitions for each set of parameters). Based on the simulation results, it has been verified that among the methods for one processor implementation, the best working methods are those of Systematic resampling, one version of Stratified resampling and Deterministic Systematic resampling. The latter method does not require drawing numbers with uniform distribution. Among resampling methods for parallel computing, best quality is characterized by two variants of stratified resampling.