Abstract
The convergence of computer vision, embedded systems, sensor networks and integrated image sensors permits the development of vision-enabled wireless sensor networks. They constitute decentralized processing systems able to realize real-time distributed vision tasks over image flows coming from different video sources. The fundamental element in these distributed systems is the smart camera. It needs to be able to locally process the captured images and to interchange data and signals with other sensory nodes of the network. It is important to understand that the result of the operation of the camera is not an image or an image flow, but a highly elaborated representation of the visual stimuli containing the relevant information about the objects in the scene and the events that take place within its visual field. This representation enables high level decision making, also at the camera node, and sharing information between nodes at a much lower communication cost.
In order to efficiently realize local processing of the visual information, we propose a vision system in which the low level tasks, are implemented by circuit structures that emulate biological vision systems. This is, by using relatively coarse processing elements, operating at the focal plane, in concurrence with image sensing. These sensor/processors work in a co-operative scheme, contain a distributed memory and exploit the physics of the devices in order to increase the efficiency without sacrificing operation speed. This type of systems, analogous to the vertebrate's retina, is now possible in CMOS technology. A system on-a-chip with this kind of sensor/processor for the more data intensive tasks and a microcontroller unit for the higher level tasks, will allow the realization of vision algorithms under tight energy constraints, thus granting vision-enabled sensor networks viable.
As for the network communication infrastructure, we will build the interface between our sensors and commercially available wireless communication modules and make use of the available networking protocols and software. Finally, we will develop a demonstrator, based on the smart image sensors conceived during this project, in order to confirm that vision-enabled wireless sensor networks constitute a viable and less costly alternative to centralized processing of images for monitoring and surveillance.