Knowledge Based Video Surveillance System

In this work, the authors propose a semantic event detection system based on a neural classifier and suitable for video surveillance. Actual video surveillance systems are composed by a distributed network of active video sensors that present their output to a human operator. The main trouble is that human operators are often unable to monitor the output of a large number of video sensors. The goal of the proposed system is to automatically collect real time information to improve the situational awareness of security providers and decision makers. Focus of the system is the evolution from classical “frame difference” paradigm to the new “known scene no alarm” and “unknown scene alarm” paradigm, where the meaning of scene is related to spatial-temporal events. The proposed system is able to detect mobile objects in the scene and to class their movements as allowed or disallowed, in order to raise an alarm, also using a video camera mounted on a motorized pan scanner. The results show that the system is able to compensate background changes due to camera motion and to detect mobile object in the scene.