Optical Acceleration for Motion Description in Videos

Modern techniques for describing motion in videos are centred around velocity descriptors based on optical flow. Realizing that acceleration is as important as velocity for describing motion information, in this paper first we propose two different algorithms to compute optical acceleration. Delving deeper into the concept of optical acceleration, we use two descriptors: histogram of optical acceleration (HOA) and histogram of spatial gradient of acceleration (HSGA), to effectively encode the motion information. To assess the effectiveness of these descriptors for motion encoding, we applied it for human action recognition and abnormal event detection in videos. In fact, we used acceleration descriptors in conjunction with velocity descriptors to get a better description of motion in videos. Our experiments reveal that acceleration descriptors could provide additional information that velocity descriptors missed and hence combining them results in a superior motion descriptor.