Robust Human Tracking to Occlusion in Crowded Scenes

Abstract

Human tracking in crowded scenes is a challenging problem because occlusion is frequently occurred. In this paper, we propose an online human tracking method which can handle occlusion effectively. Our method automatically changes a learning rate for updating tracking model according to the situation. If the tracking target is under occlusion, the learning rate decreases to reduce the influence of occlusion. However, the similarity score decreases by scale change of a tracking target as well as occlusion. To judge the occlusion or scale change, the similarity score on the Log-Polar coordinate is used. Furthermore, the size of search region is also changed according to the information about occlusion at previous frame. Experiments using the PETS2009 dataset show that our method improves tracking accuracy in crowded scenes.