Multi-pedestrian tracking based on feature learning method with lateral inhibition

As one of the hot issues in computer vision, multi-pedestrian tracking has received more and more attention recently. In this paper, under the tracking-by-detection framework, we propose a new feature learning method with lateral inhibition, combining with the traditional detection method, which is demonstrated to be effective. The tracking part utilizes a framework built upon particle filter, and the computation of the particle weight coordinately considers detector confidence, particle velocity and other factors. In addition, we carry out a procedure of particle variation before particle resampling to reduce the loss of particle diversity. As a bridge between the detector´s output and the tracker´s output, data association divides the original assignment into several independent branches for computation efficiency. Our algorithm has been shown to be feasible and effective after extensive experiments on some standard data sets.