Anomaly event detection in surveillance videos under limited supervision

Project Member(s): Chang, X.

Funding or Partner Organisation: Commonwealth Scientific and Industrial Research Organisation (Data61) (Data61 CSIRO)
Commonwealth Scientific and Industrial Research Organisation (Data61) (Data61 CSIRO)

Start year: 2022

Summary: Surveillance cameras are increasingly used in public places, e.g. streets, traffic hubs, banks, shopping malls, etc., for security concerns. However, the monitoring capability is not matched with the huge amount of surveillance videos. The result is that there is a serious deficiency in the utilization of surveillance cameras and an unworkable ratio of cameras to human monitors. For video surveillance, it is a critical task to detect abnormal events such as traffic accidents or illegal activities. Generally, abnormal events rarely occur as compared to normal activities. Therefore, to alleviate the waste of manpower and enable instant response to emergency, developing intelligent computer vision algorithms for automatic video anomaly detection is a pressing need. Events that are of interest in surveillance footages often have an extremely low probability of occurring. Most current technologies require enormous data annotation effort on each video stream prior to the deployment of the video analysis process. Also, those events labelled are based on the predefined heuristics, which makes the detection model difficult to generalize to different surveillance scenes. Thus, anomaly detection should be done with minimum supervision. Anomalies are also highly contextual, for example, wandering around in a park would be normal, but wandering around in subway platform would be an anomaly. Therefore, anomaly should be detected with contextual information involved. The research proposed will investigate a smart surveillance analysis system using big data and deep learning for decision support to public security and traffic management. It aims to timely signal an activity that deviates normal patterns and identify the time window and the spatial position of the occurring anomaly. This involves a two-stage analytical and computational framework that rigorously evaluates real-time surveillance video stream.

FOR Codes: Radio, television, film and video services, Computer vision