Change Detection From Multiple Camera Images Extended to Non-Stationary Cameras

We describe an approach for analysis of surveillance video taken from moving vehicles making repeated passes through a specific, well-defined corridor. Our goal is to detect stationary objects which have appeared in scenes along the established route. Our motivation is to address security concerns in hostile theaters where stationary surveillance cameras would be destroyed almost immediately; yet mobile camera platforms; i.e., group transport vehicles/convoys are plentiful. Challenges include illumination changes from different time/day, and handling parallax resulting from our non-stationary camera. We provide an example using artificial surveillance images taken on the Stanford University campus. Scale-up to critical security theaters would be straightforward. The approach is equally applicable to images collected by aircraft.