ANOMALY DETECTION IN SURVEILLANCE VIDEOS USING DEEP LEARNING

Deep learning has established itself as a ground breaking computing strategy in the field of computer vision. The identification of abnormalities in surveillance footage is one of the challenging cognitive tasks for which it has been frequently employed. In this instance, anomaly detection refers to the recognition of anomalous occurrences in surveillance films that may constitute security incidents or threats. For anomaly detection, deep learning approaches have outperformed more conventional approaches. Traditional closed-circuit television (CCTV) systems are inefficient and expensive because they require constant human monitoring. Thus, a system that can efficiently detect human activities in real time is required. Due to the size of a surveillance video and the time required to analyze it for activity, the video must be compressed using adaptive compression techniques. With adaptive video compression, only the areas of the video with the least emphasis are compressed; the remainder is left uncompressed. The goal of the discussion is to be able to use surveillance video to construct an automated anomalous human activity identification system that can identify unusual behaviour and alert the real-time user. The research is divided into two parts that discuss methods for adaptive video compression of surveillance films and using that compressed video as an input to identify unusual human activities.