A Poisson Multi-Bernoulli Mixture Filter for Tracking Coexisting Point/Extended Targets by Multiple Sensors' Fusion

A two step method is proposed in this paper to extend the Poisson multi-Bernoulli mixture approach original suggested in A. F. Garcia-Fernandez and Jason L. Williams's work in [3] to multiple sensors case, which are conceived to applied for both decentralized and distributed configurations. Conditions are established to determine whether tracks from different sensor fusion centers represent the same target (also named as track-to-track association). The main procedures are three folds: 1) Firstly, the PMBM filter update is derived for a generalised state and measurement model, which covered measurements originated from both point and extended targets; 2) The filtering recursion expression is obtained that propagates the Gaussian densities for point targets while gamma Gaussian inverse Wishart (GGIW for short) densities for extended targets; 3) In order to further obtain the track-to-track association problem, the S-D assignment formulation is formulated aimed to find the most likely hypothesis through solving a constrained optimization problem. The resulting filters are analysed via numerical simulations. Remarkable results can be achieved through typical swarm flocking detection illustrative examples.