Particle Filtering for Random Sets

Tracking of multiple objects simultaneously over time is an important research problem. When the number of objects to track is known, standard Bayesian methods like PDA/JPDA can be employed. However, when the number of objects to track is unknown or varies over time, tracking hypotheses with different numbers of objects have to be compared. This can be addressed in a mathematically grounded manner by viewing the set of object as a random set, in which the number of objects, N , is a stochastic variable. Tracking of random sets is formulated with finite set statistics (FISST). In this paper, we present a FISST particle filter, which is an extension of a Bayesian particle filter to incorporate the FISST formalism. Experiments show the FISST particle filter to be able to estimate both the number of tracked objects, as well as the states of the objects, robustly from noisy observations. Submitted to IEEE Transactions on Aerospace and Electronic Systems, March 2003 1