An MCMC-based particle filter for multiple target tracking

This paper applies a Markov chain Monte Carlo-based (MCMC) particle filter on the multiple target tracking problem. Traditional particle filters employ the sequential importance sampling/resampling method along with the MCMC move step, which is commonly used as a means to improve diversity among particles. The MCMC-based particle filter applied in this paper is distinct from the traditional particle filters in that: 1) It replaces the importance sampling with MCMC sampling and 2) the MCMC move is used as the sequential sampling method instead of simply as a means to improve diversity. By virtue of its information-centric property, the MCMC technique can automatically explore the posterior distribution at each sampling step. The benefit is that it allows the MCMC-based particle filter to track multiple targets without suffering from exponential complexity, which is the major drawback while using a traditional joint particle filter. Simulation results are presented in a bearings-only tracking problem for three targets and a Keplerian orbital tracking problem involving two targets.