Pedestrian Detection: The State of the Art

Опубликовано 17 августа 2016, 21:32

Pedestrian detection is a key problem in computer vision, and truly accurate pedestrian detection would have immediate and far reaching impact on areas such as robotics, surveillance, assistive technology for the visually impaired, image indexing, advanced human machine interfaces and automotive safety, among others. In the first part of the talk I will discuss the state of the art in monocular pedestrian detection, including our large-scale benchmarking effort, highlighting current successes and challenges for the research community. In the second part of the talk I will describe our own pedestrian detection approach which achieves top performance in numerous settings across multiple datasets. In particular, I will focus on a recent insight that has allowed us to perform multiscale detection at near real time on standard hardware. Our approach yields a speedup of 10-100 times over competing methods with only a minor loss in detection accuracy, and the underlying theory should be readily applicable to other domains. Finally, I will conclude by discussing open challenges in pedestrian detection and ideas on how to address them.