Coordinated Sensing in Intelligent Camera Networks

The cost and size of video sensors has led to camera networks becoming pervasive in our lives. However, the ability to analyze these images efficiently is very much a function of the quality of the acquired images. Human control of pan-tilt-zoom (PTZ) cameras is impractical and unreliable when high quality images are needed of multiple events distributed over a large area. This dissertation considers the problem of automatically controlling the fields of view of individual cameras in a camera network responsible for improving situation awareness (e.g. where and what are the critical targets and events) in a region of interest. This is achieved by understanding the performance of video analysis tasks and designing camera control strategies to improve these tasks through the quality of the source imagery. Optimization strategies, along with a distributed implementation, are proposed, and their theoretical properties analyzed. The proposed methods bring together computer vision and network control ideas. The approach introduced here consists of a system wide utility measure and a distributed optimization method. The global utility function quantifies the system performance with respect to the goals of the user. The design of several utility functions is presented such as those for improving area coverage and tracking performance. These utilities can then be combined to create a global utility measure aligned with the desired system behavior. Since exhaustive exploration of the parameter space of large camera networks is intractable, the problem is converted into a cooperative control problem where each camera optimizes local utility functions while negotiating with neighboring cameras. The result is a tractable optimization that increases the global utility until convergence at a local optima.This approach has been applied to conventional surveillance tasks, such as observing targets in a large area, as well as more complicated tasks involving competing objectives. The performance of the proposed methodologies has also been evaluated on a real life wireless network of pan-tilt-zoom (PTZ) capable cameras