Optimized imaging and target tracking within a distributed camera network

This article considers the problem of using a network of N_C dynamic pan, tilt, zoom cameras, each mounted at known and fixed locations, to track and obtain high resolution imagery for N_T(t) mobile targets each maneuvering within a confined space. The number of targets is time-varying, the targets are free to maneuver, the targets may enter or leave the region under surveillance so that N_T(t) is time-varying and may exceed N_C. Tracking a target is defined as estimating the position of the target with horizontal uncertainty less that a specified threshold P̅. Imaging a target is defined as obtaining an image with vertical resolution exceeding r̅. The problem is to organize the pan, tilt, and zoom parameters of the network of cameras at each sampling instant such that the tracking specification for all targets and the imaging specification for specific targets at times of opportunity is achieved. This problem could be addressed by centralized or decentralized methods. In this article, we are focused on distributed control of the camera network. We develop a distributed optimization solution, where we consider each camera to be an individual decision making agent. The solution involves formulation of the approach, design of a value function, and design of a probability-based camera ordering mechanism to aid convergence of the distributed network solution towards an optimal solution. Our approach is developed within a Bayesian approach to appropriately trading off value V (target tracking accuracy and target resolution) versus risk (probability of losing track of a target). This article presents the theoretical solution along with simulation results. Implementation on a camera network are in progress.