Multiple object video tracking using GRASP-MHT.

This paper presents an efficient MHT implementation, referred to as “GRASP-MHT”, featured in an integration of a greedy randomized adaptive search procedure (GRASP) and a track-oriented MHT framework. The hypothesis generating problem arising in the MHT framework is formulated as a maximum weighted independent set problem, and a GRASP module is designed to generate multiple high-quality hypotheses, thereby avoiding the need of a brute force hypothesis enumeration procedure. We validate our approach on a challenging crowd video sequence, and the experiments show the proposed method is able to capture the object trajectories successfully. Quantitative evaluation results indicate that, comparing with several well-known multitarget tracking algorithms, GRASP-MHT exhibits better performance in both data association quality and execution efficiency.