A novel algorithm to predict and detect suspicious behaviors of people at public areas for surveillave cameras

Abstract

Suspicious activities seriously endanger at public areas and personal security. There are millions of video surveillance systems used in public areas, such as streets, prisons, holy sites, airports, and supermarkets. It is essential to investigate the detection and recognition of suspicious activities contents from surveillance video. The common suspicious activities at public areas with an aspect of security are fighting, running, leave luggage and run, put an unusual packet in somewhere like a dustbin and leave. We focus on the recognition of suspicious activity and aim to find a method that can automatically detect suspicious activity using computer vision methods. Complex background, illumination changes and different distances between the human and the camera have made this topic very challenging, especially in the case of real-time applications. We adopted GMM to produce candidate regions whose has suspicious activity of motion features extracted from the magnitude information of Optical Flow, and we call this method Suspicious Activity Region Detector (SARD). Experimental results on several benchmark datasets have demonstrated the robustness of our proposed framework over the state-of-the-arts in terms of both detection accuracy and processing speed, even in crowded scenes.