Experiments with graphical model implementations of multiple target multiple Bernoulli filters

Research into random finite sets (RFSs) has provided a range of algorithms that have improved tractability over traditional alternatives due to their use of clever approximations that avoid the need for data association. This paper describes two new algorithms that were derived within the RFS framework to utilise a recently-developed approximation of the marginal data association probabilities. Experiments in a highly challenging problem involving many closely-spaced targets demonstrate the tractability of the methods, and the improvement in performance that they provide over the state-of-the-art cardinality balanced multi-target multi-Bernoulli filter.