Probabilistic multihypothesis trackerwith an evolving poisson prior

Abstract

The probabilistic multihypothesis tracker (PMHT) is an efficient multitarget tracking algorithm that performs data association under a conditional independence assumption. A key part of the measurement model is the data-association prior, which can be used as a track quality measure for track management decisions. The original PMHT makes this prior an unknown fixed parameter. The PMHT with hysteresis extended the measurement model by adding a Markov chain hyperparameter to the prior, but this came at the cost of exponential complexity in the number of targets. This complexity comes as a consequence of the normalization of the prior. This article shows that the PMHT data-association model is equivalent to assuming that targets create a Poisson-distributed number of measurements; an alternative PMHT is derived that deals directly with the Poisson model parameters and retains linear complexity in the number of targets.