Retina-Like Sensors: Motivations, Technology and Applications

Abstract

Retina-like visual sensors are characterized by space-variant resolution mimicking the distribution of photoreceptors of the human retina. These sensors, like our eyes, have a central part at highest possible resolution (called fovea) and a gradually decreasing resolution in the periphery. We will present a solid-state implementation of this concept. One attractive property of space-variant imaging is that it allows processing the whole image at frame rate while maintaining the same field of view of traditional rectangular sensors. The resolution is always maximal if the cameras are allowed to move and the fovea placed over the regions of interest. This is the case in robots with moving cameras. As an example of possible applications, we shall describe a robotic visual system exploiting two retina-like cameras and using vision to learn sensorimotor behaviors.