Gaussian sum particle filtering based on RBF neural networks

A Gaussian sum particle filter using RBF Neural Network (BRF-GSPF) is proposed to deal with nonlinear sequential Bayesian estimation. The nonlinear non-Gaussian filtering and predictive distributions are approximated as weighted Gaussian mixtures, and mixtures components are gotten by RBF neural network. This method implements conveniently in parallel way by cancelling resampling that solves weight degeneracy in particle filter. The tracking performance of the RBF-GSPF is evaluated and compared to the particle filter (PF) via simulations with heavy-tailed glint measurement noise. It is shown that the RBF-GSPF improves tracking precise and has strong adaptability.