Automated incident detection in surveillance systems

Abstract

The importance of public security cannot be overemphasized in today’s world. Surveillance systems consisting of extensive camera networks now have many applications. The London Underground has 9000 cameras, with each staff required to monitor as many as 60 at a time. This presents a significant scope for automation. Processing a sequence of images from a network presents challenges in terms of deciding upon the feature set to be extracted, and the appropriate choice of algorithm. We investigate methods based upon principal component analysis and nearest-neighbor distances to propose a system of automated intruder detection in surveillance systems and suggest new applications in Bangladesh of such systems.