Rao-Blackwellised variable rate particle filters

Variable rate particle filters have recently emerged as an alternative to multiple model techniques for tracking highly manoeuverable targets. The basic idea of variable rate methods is to apply local fits to segments of the target trajectory of variable length. Both the fit parameters and the length of the segments need to be estimated. Approximately optimal Bayesian estimation of these quantities can be performed using particle filters. In this paper a Rao-Blackwellised variable particle filter is developed which offers significant performance improvements over existing methods for a certain class of models. This is demonstrated via Monte Carlo simulations for a benchmark tracking problem.