Computer vision algorithms for automatic anomaly detection in video surveillance system

Abstract

Security has become the most essential and indispensable need across our daily life It is imperative that there is a substantial increase in the occurrence of incidents such as terror plots theft robbery fights that have disrupted normal life across the world Most of these incidents are targeted in areas where there is huge crowd gathering such as airports banks shopping malls railway stations movie theatres etc All these disrupting incidents have occurred despite the continuous monitoring of several security surveillance systems Most of the existing surveillance systems are currently used for post-mortem analysis where the video feeds are used to analyze the actions and events that resulted or triggered the disruptive event However the primary need to deploy a smart surveillance system is to perform an instantaneous real time analysis with the available video feed and immediately alert the authorities concerned to prevent any security breaches The smart surveillance system should be capable of analyzing the video feed and identifying the anomalous events automatically Anomalies are rare occurrences of certain events that are pretty unique and different from the regular pattern of events Typical examples of anomalies include presence of masked faces in public places abnormal human behaviours like sudden fighting and running and sometimes intentional abandonment of suspicious objects Most of the security breaches have occurred because of the failure to automatically detect these anomalies instantaneously and send the corresponding alert signals newline