Multi-target tracking with MCMC-based particle filters

In this article, we address the problem of multiple target tracking. Particle filter solutions must mainly cope with two problems: a high-dimensional problem and a data association problem. We propose to solve both problems simultaneously thanks to a particle filter based on a Gibbs sampler that simulates both state space and association variables. We present two possible implementations of this solution that differ in their inner structure: the first one samples the conditional densities with a Hastings-Metropolis algorithm while the second one uses importance sampling. These algorithms are shown to be as efficient as JPDA particle filters, with a dramatic reduction of the computational cost.