Statistical modeling and performance characterization of a real-time dual camera surveillance system

摘要: The engineering of computer vision systems that meet application specific computational and accuracy requirements is crucial to the deployment real-life systems. This paper illustrates how past work on a systematic methodology for performance characterization can be used develop real-time people detection zooming system given requirements. We illustrate by judiciously choosing modules performing careful analysis influence various tuning parameters it possible to: perform proper statistical inference, automatically set control quantify limits dual-camera video surveillance system. goal continuously provide high resolution zoomed-in image person's head at any location monitored area. An omni-directional camera processed detect precisely foveal camera, which has pan, tilt zoom capabilities. pan its uncertainties are shown functions underlying geometry, lighting conditions, background color/contrast, relative position person with respect both cameras as well sensor noise calibration errors. uncertainty in estimates adaptively estimate parameter guarantees user specified probability, /spl alpha/, detected face contained zoomed within image.