Wide-Area Video Surveillance
We are developing tracking and video surveillance techniques specifically designed for wide-area surveillance scenarios. As the costs of sensors decrease while their functionality and resolution increase, it is possible to imagine potentially thousands of cameras deployed over a very wide area. We are developing algorithms that allow these sensors to act together as a single logical sensor to track individuals, recognize anomalous behavior, and self-configure and calibrate.
Active projects include: multi-view geometry estimation, statistical modeling of backgrounds and foreground motion, estimation of camera relationships from observed motion (i.e.- in which sensors are tracked targets likely to emerge), and opportunistic construction of image-based models.