Attentive wide-field sensing for visual telepresence and surveillance ?

Physical and computational constraints limit the spatial resolution and field-ofview (FOV) achievable in any single sensor. In the human eye, a compromise has evolved in which resolution is high near the optical axis, falling off with eccentricity. The success of this design depends upon visual mechanisms for rapidly detecting interesting events in the periphery and motor mechanisms for accurately redirecting the fovea to peripheral targets for more detailed analysis. Prototype artificial visual systems based conceptually on these principles are being developed in our laboratory. We partition visual sensing into two components. A pre-attentive component provides a large, fixed FOV at low resolution, allowing detection of events of interest over an entire visual environment. An attentive component provides a much smaller, shiftable FOV at high resolution, designed to recognize and interpret events detected by the pre-attentive system. These prototype sensors serve as useful testbeds for computational models of visual attention, and may lead to new telepresence and surveillance technologies.