Ambient Surveillance by Probabilistic-Possibilistic Perception

Abstract

A method for quantifying ambient surveillance is presented, which is based on probabilistic-possibilistic perception. The human surveillance of a scene through observing camera sensed images on a monitor is modeled in three steps. First immersion of the observer is simulated by modeling perception of the scene from the camera locations using probabilistic perception approach. The perceptions are thereafter combined by means of probabilistic union, simulating simultaneous watching of the scene from multiple viewing positions. As third step the combined perceptions are converted to a possibility using triangular possibility density function. The latter step accounts for the fact that surveillance takes place via monitor depiction and not directly as perception of the actual physical scene. The method is described and demonstrated by means of an ambient surveillance application involving three cameras. The resulting possibility of perception is compared to the case of using two cameras, quantifying the added value of additional camera as to surveillance.