Multiview Unauthorized Human Action Recognition

This paper presents the recognition of human actions under view changes. This deploys an automotive visual surveillance system to detect abnormal behavior patterns and recognize the normal ones. If a person enters a room, video of him is captured and stored(both front view and the top view) then it is given to the training module here the video is checked if it is a normal behavior splited image is taken, whenever the action is recognized blob images are saved, and the frame counts are taken. In case, the anomaly is detected the red color will be displayed. The abnormal behavior is achieved by keep tracking the videos and blob frames and checking each frame values. Most of the current methods for action recognition are designed for limited view variations. For a given action sequence and a given type of low level features, and compute distances between extracted features for all pairs of time frames and store results in a Self-Similarity Matrix (SSM). This project approach builds upon self-similarities of action sequences over time.