Unscented Particle Implementation of Probability Hypothesis Density Filter for Multisensor Multitarget Tracking

Abstract

Probability hypotheses density (PHD) filter based on finite set statistics (FISST) is an active research area in multisenor multitarget tracking research. It estimates jointly the time varying number of targets and their states under clutter environment, and doesn’t need data association for multitarget tracking, which breaks through traditional tracking methods. Two kinds of implementation algorithms of this technique have been developed: sequential Monte Carlo PHD (SMCPHD) filter and Gaussian mixture PHD (GMPHD) filter. However, the latter is intractable for nonlinear non-Gaussian tracking models, while the former is equivalent of the particle filter known to be inefficient. Based on the ideas from unscented particle filter (UPF), we present an unscented particle implementation of PHD filter to enhance its efficiency, which is compared with SMCPHD algorithm by the experimental simulation. It is showed that presented implementation is more effective in the tracking accuracy, and has a better ability of state estimation.