Improved GRASP-MHT for Possibly Unresolved Measurements

Abstract

The multiple hypothesis tracker (MHT) generally assume that each measurement originates from one target independently, but sensors with finite resolution can only generate one unresolved measurement for multiple targets when they are in close proximity. The possible presence of unresolved measurements complicates the data association problem and significantly degrades the performance of the tracking algorithm. An improved greedy randomized adaptive search procedure MHT (GRASP-MHT) algorithm is proposed to handle the situation of unresolved measurements. In this algorithm, the scores of track hypothesis associated with unresolved measurement are calculated, and the data association problem of unresolved measurements is modeled as a maximum weight independent set problem (MWISP). Simulation results show that the improved GRASP-MHT can achieve a better tracking performance in the presence of unresolved measurement and retains most of the advantages of GRASP-MHT.