Tracking and recognizing two-person interactions in outdoor image sequences

This paper presents a methodology for tracking persons and identifying two-person interactions in outdoor image sequences. By locating and tracking two persons over consecutive frames of monocular grayscale image sequences, we classify their interactions into several classes. Some of the interaction classes are: one person leaves another stationary person; two persons meet coming from different directions; and one stationary person starts following another walking person. We use side-view image sequences obtained by a fixed camera. In these image sequences the subjects are frequently occluded and move perpendicular to the direction of the camera. We present results on tracking of persons and their interactions.