A Belief Space Perspective of RFS based Multi-Target Tracking and its Relationship to MHT

Abstract: In this paper, we establish a connection between Reid{'}s HOMHT and the modern Random Finite Set (RFS)/ Finite Set Statistics (FISST) based methods for Multi-Target Tracking. We start with an RFS description of the Multi-Target probability density function (MT-pdf), and derive the prediction, and update equations of the MT-tracking problem in the RFS framework from a belief space perspective. We show that the RFS pdf has a hypothesis dependent structure that is similar to the HOMHT hypotheses structure. In particular, we examine the different hypotheses, and derive the hypotheses update equations under the FISST recursions, and clearly show its relationship to the classical HOMHT hypotheses and hypothesis weight update formula, thereby establishing a connection between the methods.