The Gaussian Particle multi-target multi-Bernoulli filter

Multi-target multi-Bernoulli (MeMBer) filter is a new attractive approach to tracking an unknown and time-varying number of targets. In this paper, we present a new implementation of the MeMBer recursion-the Gaussian particle MeMBer (GP-MeMBer) filter-for nonlinear models. The probability density in the multi-Bernoulli is approximated by a weighted sum of Gaussians, as in the existed Gaussian mixture (GM-MeMBer) filter, but the target dynamics or observation can be nonlinear. Monte Carlo integration is applied for approximating the prediction and posterior densities in the multi-Bernoulli and the multi-Bernoulli existence probability. The simulation results verify the effectiveness of the proposed GP-MeMBer filter.