Biologically inspired vision sensors

Abstract

The paper starts with a brief discussion of the biological vision system that serves as a model for the three types of vision sensors described later on. A retina-like CCD sensor whose computational properties are embedded in its structure is described first, followed by a CMOS tracking sensor that consists of a fovea for smooth pursuit and a periphery for saccadic motion control. This sensor incorporates logarithmic compression, edge detection, direction-of-motion detection and centroid localization. Finally, a CMOS sensor for the detection of image features is discussed. The sensor extracts lower-level features, such as line orientation, line stops, and intersections, in a hierarchical fashion, similar to what the simple and complex cells do in the biological system.