Tracking Groups of People in Video with Histogram-PMHT

The Histogram Probabilistic Multi-Hypothesis Tracker (H-PMHT) algorithm is a parametric track-before-detect method. It locates targets in imagery by fitting a mixture of probability densities. In recent work, an unknown target shape was accommodated through the use of a Wishart prior on the measurement probability density, this version of the algorithm is referred to as H-PMHT-RM. This paper applies H-PMHT-RM to video surveillance of people and extends the track management model to incorporate the merging and splitting of targets. The performance of the algorithm is compared with a particle filter based method on benchmark data from the CAVIAR project.