Automatic detection of situations for intervention using 3D image reconstruction and template matching

Due to the public transportation companies' need to ensure the safety and security of the passengers, the recent years have witnessed a substantial increase in the number of surveillance cameras installed in these stations. Regarding to this increasing number, the companies automatically face the problem that human operators have some difficulties thoroughly monitoring all cameras' outputs meticulously. In this paper, we will present a system that will monitor a selected zone in a public station and will serve solely as an assisting component for the surveillance system operator who will take the final appropriate measures for intervention. The system will be analysing the scenes in real time by reconstructing the foreground objects in 3D space and using a series of image processing methods to detect a predefined "Situation in Need of Intervention" (SNI). This paper will describe three main SNIs (falling/lying on the ground, act of human aggression and unattended objects) and will show the methodology that was used to classify a scene under these three SNIs.