Monocular-Vision-Based Mobile Robot Global Localization

Abstract

An environmental map built with monocular vision is used to implement mobile robot global localization.The feature matching is implemented with the KD-treebased nearest search approach.The features are extracted with Scale Invariant Feature Transform(SIFT),and discribed with highly distinctive multi-dimensional vector,making features be invariant to changes in illumination,scale,3D viewpoint and noise.A robust localization based on RANSAC(RANdom SAmple Consensus) approach is presented.Experiments on robot Pioneer 3 with monocular CCD camera in our real indoor environment show that our method is of high precision and stability.