A Critical Look at the PMHT

We combine concepts from numerous papers to provide a derivation and description of a generalized Probabilistic Multi-Hypothesis Tracker that can track multiple targets in a cluttered environment, utilizing multiple sensors and feature measurements, if available. Additionally, we provide a full derivation of the algorithm, including parts omitted or abbreviated in other work. We also provide an improved analytic solution for the proir target-measurement probabilities (the πs) conditioned on the number of observations, a simplified method of performing the maximization step of the algorithm when multiple sensors are used, a consistent covariance approximation of the algorithm when using multiple sensors, explore the use of deterministic annealing to improve performance and discuss implementation difficulties. Index Terms Tracking, PMHT, EM Algorithm, Fusion 1 OVERVIEW Since its creation by Streit and Luginbuhl in 1993 [60], much research has been done on the Probabilistic Multi-Hypothesis Tracker (PMHT). In this paper, we combine concepts from past works and provide a general version of the PMHT algorithm allowing for tracking in the presence of clutter (false alarms) and missed detections and the utilization of classification data, range rate information and multiple synchronous sensors. This version makes no changes to the basis of the original algorithm, which is the Expectation Maximization (EM) algorithm. As a result, this generalized PMHT algorithm may be used as an improved foundation for other versions of the PMHT that build upon or alter the basis of the algorithm, such as the Multi-Frame Assignment PMHT (MFPMHT) accounting for missed detections by Blanding, Willett, Streit, and Dunham [7]. Being a generalized version of the PMHT, the algorithm Manuscript received April 3, 2009. This research was supported by the Office of Naval Research under contract N00014-07-1-0429, and by Vectraxx, Inc. All authors are with the Department of Electrical and Computer Engineering, University of Connecticut, 371 Fairfield Way, U-2157, Storrs, Connecticut 06269 USA (e-mail: crouse, marco.guerriero, [email protected]) We would like to thank Yaakov Bar-Shalom for his comments.