A VOTING STRATEGY FOR HIGH SPEED STEREO MATCHING

Abstract

In this paper we propose a new stereo matching algorithm for real-time obstacle detection in front of a moving vehicle. The stereo matching problem is viewed as a constraint satisfaction problem where the objective is to highlight a solution for which the matches are as compatible as possible with respect to specific constraints. These constraints are of two types: local constraints, namely position, slope and gradient magnitude constraints, and global ones, namely uniqueness, ordering and smoothness constraints. The position and slope constraints are first used to discard impossible matches. Based on the global constraints, a voting stereo matching procedure is then achieved to calculate the scores of the possible matches. These scores are then weighted by means of the gradient magnitude constraint. The correct matches are finally obtained by selecting the pairs for which the weighted scores are maximum. The performance of the voting stereo matching algorithm is evaluated for real-time obstacle detection using linear cameras.