An improved multiple object tracking algorithm of GMMCP

In this paper, we proposed an improved GMMCP algorithm for multiple object tracking (MOT) problems. Compared with the original algorithm, we add a real-node item into the objective function model, using the detection confidence information to further rich the original model, and we add the fourth constraint to guarantee the calculation. The same as the original algorithm, we use CPLEX toolkit to solve the Mixed Binary-Integer Program (MBIP) problem and obtain the optimal solution. Three datasets are tested in our experiment including Parking-Lot 2, TUD-Crossing and TUD-Stadtmitte. We evaluate our method by CLEAR MOT metrics and compare our results with several state-of-art methods. The quantitative results show that our method performs better to varying degree in all the image sequences than the original method.