Application of proximal support vector regression to particle filter

Abstract

An improved particle filter for nonlinear, non-Gaussian estimation is proposed in this paper. The algorithm consists of a particle filter that uses a proximal support vector regression (PSVR) based re-weighting scheme to re-approximate the posterior density and avoid sample impoverishment. A regression function is obtained by PSVR over the weighted sample set and each sample is re-weighted via this function. Then, posterior density of the state is re-approximated to maintain the effectiveness and diversity of samples. Two experimental results demonstrate that the efficiency of the proposed algorithm compared with the generic particle filter and Markov Chain Monte Carlo (MCMC) particle filter.