Multiple object tracking based on modified algorithm of GMMCP tracker

We formulate multiple object tracking (MOT) as a Generalized Maximum Multi Clique problem (GMMCP) without any simplified version in problem formulation or optimization, which makes full use of image evidence. Afterwards, we solve the GMMCP through Mixed Binary-Integer Program (MBIP) and obtain the final motion trajectories of the multiple objects. In order to improve the algorithm's performance, we employ Aggregated Dummy Nodes (ADN) and modify the feature of object appearance. We performed experiments and evaluated our algorithm on three video sequences including TUD-Stadtmitte, TUD-Crossing and Parking-Lot 2. We compared our method with some state-of-art methods and showed that the algorithm we proposed performs an excellent precision in multiple object tracking.