A comparison of hardware resources required by real-time stereo dense algorithms

Abstract

Many algorithms for computing correlation based stereo correspondence have been proposed. Some of them can be implemented on specialized architectures, in order to obtain results in real time. In this communication, we propose an experimental comparison of the amount of hardware resources required for implementing these algorithms. An efficient architecture is presented, the STREAM, (acronym for systeme temps-reel d'extraction et d'analyse du mouvement, i.e. real-time motion extraction and analysis system) which is a processor dedicated to image sequence analysis