Hybrid Particle Filtering Based on an Elitist Resampling Scheme
Mhd Modar Halimeh, Christian Huemmer, Andreas Brendel, Walter Kellermann · 2018
In this paper, the elitist resampling particle filter (ERPF) is introduced. The ERPF is an approach to combine generic particle filters, i.e., the SIS and SIR particle filter, based on an evolutionary selection of particles, which introduces a longterm memory to the selected group of the particles. Thereby, the ERPF aims at fusing the advantages, e.g., robustness and computational efficiency, and mitigating the drawbacks of each filter, e.g., degeneracy and sample impoverishment. Two variants of the ERPF are presented in this paper: the discrete ERPF (DERPF) and the continuous ERPF (CERPF), which to the authors' knowledge represent new forms of particle filters. The two proposed hybrids are compared to the corresponding original filters using two well-established benchmark models to illustrate that the evolutionary combinations provide a better estimation accuracy and exhibit an increased robustness against outliers.