The Rao-Blackwellized marginal M-SMC filter for Bayesian multi-target tracking and labelling

In multi-target tracking (MTT), we are often interested not only in finding the position of the objects, but also allowing individual objects to be uniquely identified with the passage of time, by placing a label on each track. In some situations, however, observability conditions do not allow us to maintain the consistency in the correspondence between track labels and true objects. In this situation, it may be useful for the operator to know the probability of loss of this consistency, i.e. the probability of labelling error. This is theoretically possible using Bayesian multi-target tracking approaches like the Multi-target Sequential Monte Carlo (M-SMC) and the Multiple Hypothesis Tracking (MHT) filters, but unfortunately, it is well-known that these methods suffer from a form of degeneracy known as “self-resolving‿, that causes the probability of labelling error to be severely underestimated. In this paper, we propose a new Sequential Monte Carlo algorithm for the multi-target tracking and labelling (MTTL) problem, the Rao-Blackwellized marginal M-SMC filter, that deals with self-resolving and is valid for multi-target scenarios with unknown/varying number of targets.