Pedestrian detection using 3D optical flow sequences for a mobile robot

In this paper, we propose a method to detect pedestrians using 3D optical flow sequence in dynamic environments captured by a stereo camera mounted on a mobile robot. First, 3D optical flow by combining range and intensity image sequences of a stereo camera is generated, and connected components in range image are segmented as objects by using range image. Second, flow vector variance among segmented object regions is calculated to find out whether a pedestrian is in single or double support phase. Flow vector variance characteristic of single or double support phases in walking is obtained by analyzing clinical walking data. Third, time series of this matching result are used to detect pedestrians. Since 3D optical flow vectors are invariant in a dynamic environment including camera motion, this method is applicable for a mobile robot. We implemented this system and examined the performance on our mobile robot Pen2.