Genetic algorithms applied to binocular stereovision

Abstract

This paper describes an original approach to the problem of edge-based binocular stereovision. The tokens to be matched are subchains of the chains of connected pixels. Local constraints of the stereovision problem are first used in associating to each token a set of potential matches. Global constraints are embedded in a cost function and we look for the minimum of this cost function. The optimisation search is conducted using genetic algorithms.