Histogram PMHT with target extent estimates based on random matrices

Abstract

Conventional tracking approaches are based on the assumption that the targets to be tracked are point targets and that the measurements to be processed are point measurements. However, when a sensor provides image data of high resolution in which targets might be distributed over several display cells, neither assumption is suitable. In such applications the estimation of the target extent and the utilization of the complete image frame are crucial to achieve good tracking performance. Recently, a Bayesian filter for single extended object tracking based on random matrices has been proposed. In this approach ellipsoidal object extents are modeled by random matrices and treated as additional state variables. This article deals with the integration of random matrices into the Histogram Probabilistic Multi-Hypothesis Tracker. The novel approach tracks multiple extended targets directly in an image sequence without previous point measurement extraction. The superiority of the algorithm is proven by simulations.