Localization of Strangeness for Real Time Video in Crowd Activity Using Optical Flow and Entropy

Abstract

Anomaly detection, which is also referred to as novelty detection or outlier detection, is process of identifying unusual occurrences, observations, or events which considerably differ from the bulk of data and do not fit a predetermined definition of typical behavior. Medicine, cybersecurity, statistics, machine vision, law enforcement, neurology, and financial fraud are just a handful of the industries where anomaly detection is used. In the presented study, an online tool is utilized to identify crowd distortions, which could be brought on by panic. An activity map is produced with the use of numerous frames to show the continuity regarding the flow over time following the global optical flow has been calculated in the quickest time and with the highest precision possible utilizing the Farneback approach to calculate the magnitudes. Utilizing a specific threshold, the oddity in the video will be picked up by the activity map's generation of an entropy. The results indicate that the maximum entropy level for indoor video is <0.16 and the maximum entropy level for outdoor video is >0.45. A threshold of 0.04 is used to determine whether a frame is abnormal or normal.