Histogram-PMHT for extended targets and target groups in images

Abstract

This article deals with the integration of random matrices into the Histogram Probabilistic Multi-Hypothesis Tracker (H-PMHT), a parametric track-before-detect method that locates targets in imagery by fitting a mixture of probability densities. The random matrices are used to describe the unknown physical extent of targets in the sensor image, a parameter that can change with time depending on the target orientation and sensor geometry. The track management model is extended to allow merging and splitting targets. The performance of the algorithm is quantified through simulations and using a benchmark people surveillance data set from the CAVIAR project.