Multiple model sequential MCMC for jump Markov systems

The problem of discerning between multiple models arises in several domains and applications. In general the use of models allows an improved and more robust processing of the data, when there is a good match between the model and the data. Consequently correct model selection is essential. New algorithms are thus investigated here which can cope with multiple models. In particular this is shown for the example of tracking of targets presenting sudden maneuver. Particle filter (PF) techniques based on the Interacting Population Markov Chain Monte Carlo (IP-MCMC) scheme present more degrees of freedom in algorithm design with respect to classical Sampling importance resampling (SIR) PF. In this paper two families of IP-MCMC PF algorithms are proposed, respectively Interacting Multiple Models IP-MCMC (IMM-IP-MCMC) and Multiple Model Selection IP Reversible Jump MCMC (MMS-IP-RJMCMC), to tackle the problem of motion model selection for highly maneuverable targets and compared to the Interacting Multiple Models SIR (IMM-SIR) PF implementation. Simulations demonstrate that the proposed algorithms yield a more robust performance than the SIR implementation.