Video Behavior Analysis Using Topic Models and Rough Sets [Applications Notes]

Liang Zhao, Lin Shang, Yang Gao, Yubin Yang, Xiuyi Jia

Abstract: Video behavior analysis is both meaningful and challenging in computer vision. As the monitoring equipment become readily cheaper and available, massive video data have been generated. As abnormal behaviors are rare, manual surveillance may drastically increase the chances of false positives or false negatives. It is hence useful to achieve automatic surveillance. Abnormal behaviors are usually complex and short. It is however difficult to completely list the abnormal behaviors. We can view abnormalities as outliers. Sometimes abnormalities co-occur with normalities, which cannot be distinguished in some feature space.