Information-theoretic Surveillance

left camera

right camera

Collaborative tracking of targets by multiple cameras with information-theoretic utility functions. Using Entropy and Mutual Information to encode the objective in the surveillance allows a mixture of different, possibly conflicting objectives. In my recently published work (M2SFA2, CVPR2008), active (pan, tilt, zoom) cameras are controlled according to two aims: tracking performance and acquisition performance. These objectives contradict if the cameras cannot supervise the whole area at the same time.