A Robust NF-GMMCP algorithm for Multi-Pedestrian Tracking

摘要

In this paper, we proposed a novel NF-GMMCP algorithm for Multi-Pedestrian Tracking (MPT) problems, which implements data association in video sequences by connecting network flows(NF) and maximum generalized multi-group algorithm (GMMCP). In the case where the detection set of the video sequence is given, the first step is to generate the pre-level tracklets by using dynamic programming (DP) algorithms to compute shortest paths in the NF problems. Secondly, to find the middle-level tracklets by using CPLEX toolkit to solve the Mixed Binary-Integer Program (MBIP) problem of GMMCP algorithm. Finally, we selected a simple threshold determination model to create final trajectories. CLEAR MOT metrics are used to evaluate our model performance. The experimental results on three challenging video sequences show that our method is able to obtain the superior performance under the compare with recently several state-of-the-art methods.