Detection and Localization of Anomalies from Videos based on Optical flow Magnitude and Direction

摘要

Anomalies in video scenes means unexpected or unusual activity which is usually not frequently observed. Such activities hence are rare and require sudden attention so that it can be detected as early as possible. There is a need to automatically identify and locate where such anomaly is present. Optical flow magnitude and direction based method is an automated system built on motion, position and statistical features of moving objects present in video. Moving objects are identified by means of optical flow and are represented using bounding box. The normal behaviors is learned beforehand for different objects. A generalization of normal behavior is captured by clustering different directional motions in the scene. Anomalous behavior of objects are detected and localized using motion and positions differing from normal behavior. The performance of proposed method is compared with existing methods by using standard benchmark datasets available online such as UCSD and UMN.