Detection, classification, and collaborative tracking of multiple targets using video sensors

The study of collaborative, distributed, real-time sensor networks is an emerging research area. Such networks are expected to play an essential role in a number of applications such as, surveillance and tracking of vehicles in the battlefield of the future. This paper proposes an approach to detect and classify multiple targets, and collaboratively track their position and velocity utilizing video cameras. Arbitrarily placed cameras collaboratively perform self-calibration and provide complete battlefield coverage. If some of the cameras are equipped with a GPS system, they are able to metrically reconstruct the scene and determine the absolute coordinates of the tracked targets. A background subtraction scheme combined with a Markov random field based approach is used to detect the target even when it becomes stationary. Targets are continuously tracked using a distributed Kalman filter approach. As the targets move the coverage is handed over to the "best" neighboring cluster of sensors. This paper demonstrates the potential for the development of distributed optical sensor networks and addresses problems and tradeoffs associated with this particular implementation.