Human identification by gait from event-based camera

Gait recognition is a computer vision problem complementing other identification problems such as face or iris recognition. Unlike other identifiers, gait is based on the motion of body points and thus it can be captured by Dynamic Vision Sensor (DVS). In this work, we explore the possibility of gait recognition in the event stream by visualizing it and applying the existing method which achieves state-of-the-art results on several benchmarks. During the investigations several related problems such as moving object detection and human pose estimation are considered as auxiliary ones. Several algorithm settings are evaluated and compared. The obtained results show that all the problems can be solved in event-based data with high quality, which is close to the quality achieved on conventional colored videos.