Evaluation of a probabilistic multihypothesis tracking algorithm in cluttered environments

This research examines the probabilistic multihypothesis tracker (PMHT), a batch-mode, empirical, Bayesian data association and tracking algorithm. Like a traditional multihypothesis tracker (MHT), track estimation is deferred until more conclusive data is gathered. However, unlike a traditional algorithm, PMHT does not attempt to enumerate all possible combinations of feasible data association links, but uses a probabilistic structure derived using expectation-maximization. This study focuses on two issues: the behavior of the PMHT algorithm in clutter and algorithm initialization in clutter. We also compare the performance between this algorithm and other algorithms, including a nearest neighbor tracker, a probabilistic data association filter (PDAF), and a traditional measurement-oriented MHT algorithm.