Mixed Particle Filtering for Maneuvering Target Tracking in Clutter

Abstract

The particle filtering (PF) is a recursive sub-optimal Bayesian estimator. The multiple model particle filtering (MMPF) has been proposed for tracking a maneuvering target. In a cluttered environment, probabilistic data association (PDA) is incorporated into MMPF to overcome the measurement-origin uncertainty. While the particle filtering is fairly easy to implement, its main drawback is that it is quite computation intensive, with the computation complexity increasing quickly with the state dimension. Rao-Blackwellized PF or marginalized PF is one remedy to this problem by marginalizing out the state appearing linearly in the dynamics. In this paper, we introduce the mixed particle filtering PDA (MPF-PDA) algorithm, an efficient variant on the PF for nonlinear maneuvering target tracking in clutter. Each particle samples a discrete mode and approximates the continuous state by a Gaussian distribution which is updated by a combination of the unscented Kalman filter (UKF) and PDA. The discrete mode is estimated by an improved PF combined with PDA. The posterior distribution of the target state is approximated with a mixture of Gaussians. Monte Carlo simulations show performance improvement of the proposed algorithm over traditional bootstrap particle filtering, and the superiority for large clutter densities.