Tracking multiple pedestrians through detection failures in videos
Zhengqiang Jiang, Benlian Xu · 2015
This paper presents a method that integrates an effective data association with tracking persistency constraint into motion models to track pedestrians in video sequences captured by a fixed camera. Pedestrians are detected at each frame using the HOG+B/F human detector which combines HOG human detector and background subtraction technique. We impose the tracking persistency constraint into data association to avoid incorrect data association due to detection errors of the pedestrian detector. Our experimental results show that our tracking method outperforms one that uses data association without tracking persistency constraint and can handle partial occlusion.