MAC-RANSAC: a robust algorithm for the recognition of multiple objects

International audienceThis paper addresses the problem of recognizing multiple rigid objects that are common to two images. We propose a generic algorithm that allows to simultaneously decide if one or several objects are common to the two images and to estimate the corresponding geometric transformations. The considered transformations include similarities, homographies and epipolar geometry. We first propose a generalization of an "a contrario" formulation of the RANSAC algorithm proposed in [Moisan and Stival 04]. We then introduce an algorithm for the detection of multiple transformations between images and show its efficiency on various experiments