Virtual camera modeling for multi-view simulation of surveillance scenes

Abstract

A recent trend in research is to leverage on advanced simulation frameworks for the implementation and validation of video surveillance and ambient intelligence algorithms. However, in order to guarantee a seamless transferability between the virtual and real worlds, the simulator is required to represent the real-world target scenario in the best way possible. This includes on the one hand the appearance of the scene and the motion of objects, and, on the other hand, it should be accurate with respect to the sensing equipment that will be used in the acquisition phase. This paper focuses on the latter problem related to camera modeling and control, discussing how noise and distortions can be handled, and implementing an engine for camera motion control in terms of pan, tilt, and zoom, with particular attention to the video surveillance scenario.