Specialized hardware for real time navigation

Abstract

The goal of this work is to propose a specialized hardware architecture to permit a real time visual navigation of a mobile robot. The navigation method is a two-stage approach to extract features and to match them over an image sequence acquired during a mobile robot motion in order to recover a 2D motion field useful to extract informations about ego-motion parameters. This paper describes an hardware implementation of the first stage of the method consisting in feature extraction and raw match computation through radiometric measure minimization. Performances of our hardware permits the whole system to reach the real time by requiring a processing rate of 40 MHz.