Smart camera networks are well suited for a broad range of applications. Techniques from multiple disciplines need to be brought together to effectively map an application onto the energy, processing, and communication constraints of the camera nodes. For this reason, we propose an application-driven design methodology that enables the determination of an output set of operation parameters given an input set of application requirements. This design methodology, for instance, can yield the camera resolution and node placement necessary to achieve accuracy performance and network lifetime targets. As the underlying estimation technique used in various applications, we adapt distributed Bayesian filtering, which fuses node observations in a probabilistic manner. We will demonstrate how this approach can readily enable key network tasks like target tracking, occupancy sensing, and contour tracking in a collaborative multi-camera setting. Observation models for monocular and stereoscopic vision systems are derived that link the physical world through an analytical formulation to a simulation model. The observation models were experimentally validated with MeshEye, our custom designed smart camera mote. Its novel hybrid-resolution vision system deploys a pair of kilopixel image sensors to trigger a higher resolution sensor and guide the region of activity in its field of view.;Our contribution lies in combining the characteristics of distributed Bayesian estimation and camera mote design to devise a methodology for application-driven design of smart camera networks. This includes case studies that explore several fundamental dependencies between operation parameters and performance metrics. A design example of indoor target tracking with occlusions present illustrates our methodology by jointly optimizing network topology, camera resolution, and sensor selection.
展开▼
