Investigation on an analog stereo retina for automobile applications

Abstract

This paper presents an investigation on analog implementation of a stereovision algorithm suited to obstacle detection applications, especially to future intelligent vehicles or autonomous robots. The stereo algorithm is derived from the second algorithm of Marr (1979) by using local extrema extracted from DoG filtered images as matching primitives, instead of zero-crossing points. The DoG filter is realized by a extremely simple and precise capacitive IIR 2D spatial Gaussian filter, using only C and switching MOS-the most precise elements in CMOS technology. This capacitive network is much more precise than other implementations and simplifies considerably the algorithm-hardware conversion. The matching primitives extraction is implemented by using charge sharing and voltage comparison, no subtraction operation is needed. This stereo retina has been simulated with real-scene images taken from a running car on highway. The simulation results are very positive and a 128-pixel line retina has been designed and submitted to fabrication.