Multiple target tracking with a pixelized sensor

This paper describes a computationally efficient method for tracking multiple moving targets. The method is predicated on estimation of the joint multitarget probability density (JMPD), which is a single probabilistic entity capturing uncertainty in both the number of targets and the states of the individual targets. The non-Gaussian/nonlinear measurement model adopted here does not permit exact computation of the JMPD so some method of approximation is required. A novel particle filtering algorithm for recursive estimation of the JMPD is proposed which provides computational tractability by automatically factoring the high dimensional multitarget state when applicable and using the optimal importance density with Rao-Blackwellisization on each of the factors. The efficiency of the proposed algorithm is shown via simulation results using real, recorded target trajectories.