Sequential Monte Carlo implementation of the track-oriented marginal multi-Bernoulli/poisson filter

The TOMB/P filter [Williams, 2011, 2015] is an attractive method for multiobject tracking. However, its original formulation is computationally feasible only for linear-Gaussian system models, and it suffers from the track coalescence effect. Here, we propose a sequential Monte Carlo (SMC) implementation of the TOMB/P filter, termed the TOMB/P-SMC filter, which avoids these drawbacks. We demonstrate the performance of the TOMB/P-SMC filter in a challenging scenario with a nonlinear range-bearing measurement model, low probability of detection, strong clutter, and intersecting objects. It is observed that track coalescence is significantly reduced, and that the TOMB/P-SMC filter is able to outperform SMC implementations of previously proposed filters such as the cardinalized PHD filter and the cardinality-balanced multi-Bernoulli filter.