Computing lightness constancy with an APS-based silicon retina

Abstract

A silicon retina is an intelligent vision sensor that can execute real-time image pre-processing by using a parallel analog circuit that mimics the structure of the neuronal circuits in a vertebrate retina. In order to enhance robustness against changes in lighting conditions, we designed and fabricated a frame-based, wide dynamic range silicon retina with an active pixel sensor that approximates the logarithmic illumination-to-voltage transfer characteristics by using a time-dependent stepped reset voltage technique. The chip in this study realized dynamic range wide enough for perceiving objects in both indoor and outdoor environments. Moreover, the combination of the logarithmic-like photosensor and the Laplacian-Gaussian-like filtering by a resistive network in the chip produces the response with lightness constancy, that is, the response of the chip depends only on the contrast of the reflectance of objects, and not on the changes in illumination. The present silicon retina is suitable for the front end of the vision system in autonomous mobile robots in the real world.