This paper introduces a novel approach to Bayesian tracking in which the posterior distribution is represented by a random sample or ensemble of possible target states. Such a representation may be viewed as a special case of the traditional particle filter approach wherein all particles maintain strictly uniform weights. Measurement updates are performed using a Markov Chain Monte Carlo technique which has been adapted to use multiple chains and a variable pseudo temperature akin to that used in simulated annealing. The general formalism is illustrated in an example of sonar-based target tracking for antisubmarine warfare. For this example, specific motion models and likelihood functions are developed for both target and clutter hypotheses. The technique is examined in the context of results from a recent multistatic seatrial using an echo repeater target and compared against those of a traditional particle filter.

This content is only available via PDF.