Learning proactive control strategies for PTZ cameras

This paper introduces a camera network capable of automatically learning proactive control strategies that enable a set of active pan/tilt/zoom (PTZ) cameras, supported by wide-FOV passive cameras, to provide persistent coverage of the scene. When a situation is encountered for the first time, a reasoning module performs PTZ camera assignments and handoffs. The results of this reasoning exercise are 1) generalized so as to be applicable to many other similar situations and 2) stored in a production system for later use. When a “similar” situation is encountered in the future, the production-system reacts instinctively and performs camera assignments and handoffs, bypassing the reasoning module. Over time the proposed camera network reduces its reliance on the reasoning module to perform camera assignments and handoffs, consequently becoming more responsive and computationally efficient.