Pedestrian Tracking-by-Detection Using Image Density Projections and Particle Filters

Video-based people detection and tracking is an important task for a wide variety of applications concerning computer vision systems. In this work, we propose a pedestrian tracking-by-detection system focused on the role of computational performance. To this aim, we have developed a computationally efficient method for people detection, based on background subtraction and image density projections. Tracking is performed by a set of trackers based on particle filters that are properly associated with detections. We test our system on different well-known benchmark datasets. Experimental results reveal that the proposed method is efficient and effective. Specifically, it obtains a processing rate of 22 frames per second on average when tracking a maximum number of 9 people.