A Real-Time Smart Sensor System for Visual Motion Estimation

Abstract

This paper presents a new intelligent sensor system for optical flow estimation. The sensor provides a field of 128 x 128 velocity vectors. A prototype system is fabricated consisting of a CMOS photo diode matrix serving as front, a digital CMOS ASIC as core of the system, and an interface to high level processing. In contrast to conventional computer vision systems, which process image data frame by frame, our system processes image data pixel by pixel synchronously with reading the photodiode matrix. This pipelined processing architecture enables the sensor system to process up to 50 frames/second. The optical flow sensor is intended to be used on an autonomous vehicle. In combination with a DSP-board for high-level image processing tasks, the sensor will be used for secure navigation and dynamical collision avoidance. 1 Introduction So far many approaches exist, that use image sequence processing on autonomous mobile robots [21]. Typical tasks are the detection of either static or ...