An Analog Visual Feature Extraction Retina for Real-Time Stereo Vision

In this paper, we present an analog VLSI spatio-frequency analysis based visual feature extraction retina, dedicated to the real-time stereo vision. In this retina, local extreme points in the DoG filtered image are extracted as pertinent visual features. A 128-pixel line-based prototype chip is presented with experiment results. A processing speed of 100k lines/s has been obtained (exposition time excluded), it will be used in an integrated analog stereo vision system for real-time obstacle detection task.