Development of video analytic Algorithm for human behavior Recognition

Abstract

During the last decade, there has been a massive increase in the newlineusage of video cameras by both public and private agencies to capture and newlinerecord various happenings such as accidents, crimes, suspicious activities, newlineterrorism and vandalism, thereby helping the law enforcing authority to newlineproduce valid evidence against various crimes in the court of law. This raises newlinea gamut of research challenges with hard-hitting economic impact to capture, newlinestore, and distribute video, while leaving the task of threat detection newlineexclusively to human operators. Video Analytics are designed to assist newlinesecurity professionals in making quicker decisions while responding to newlinesuspicious human behavior. Thus, human behaviour analysis is a crucial part newlinein Video Analytics that starts with extracting visual cues and primitive events. newlineThese extracted cues are important in detecting complex behavioral patterns. newlineThe heterogeneous non-verbal behavioral cues like emotions, gaze, gesture, newlineposture and gait are major components to perceive an assailant. newlineTo extract the non-verbal cues for content understanding, the input newlinevideo is classified into shot type like Long Shot, Medium Shot, and Close up newlineshot. An effective shot classification method based on transfer learning for newlinesurveillance videos is proposed. A close-up shot highlights the emotional state newlineof the person caught in the camera; medium Shot conveys less emotional newlinecontent with some information regarding the location where the action is newlinetaking place, while the Long Shot emphasizes more on the context thus newlinehelping in extracting the details of the location, multiple person interaction, newlinesocial interaction etc. newline