PhD forum: Probabilistic surveillance with multiple active cameras

Abstract

This work presents a method to control multiple, but diverse pan-tilt-zoom cameras which are sharing overlapping views of the same spatial location for the purpose of observation of this scene. We cast this control input selection problem in an information-theoretic framework, where we maximise the expected mutual information gain in the scene model with respect to the observation parameters. Overall this yields a framework in which heterogeneous active camera types can be integrated cleanly and consistently, obviating the need for a wide-angle supervisor camera or other artificial restrictions on the camera parameter settings.