A multi‐stage matching algorithm for mobile robot navigation

Abstract

Purpose Stereo vision is an attractive perception technique for mobile robots navigation. Stereo matching is a crucial part of stereo vision and its precision dominates the precision of reconstruction. Based on a geometry constraint applicable to natural terrain, the purpose of this paper is to present a multi‐stage stereo matching algorithm to improve matching accuracy. Design/methodology/approach In the multi‐stage matching algorithm, points with larger intensity gradient are matched in earlier stages. Using several constraints and statistical means, information from earlier stages is utilized to assist in matching of later stages to improve matching accuracy. Findings The multi‐stage matching algorithm improves the matching accuracy of stereo pairs of natural terrain in various conditions. Research limitations/implications The algorithm demonstrates advantages over area‐matching algorithm both in matching accuracy and computation efficiency. However, if used for real‐time navigation, it still needs the assistance of specialized hardware or window selection technique. Practical implications The algorithm is able to produce dense disparity maps of natural terrain with fairly high accuracy and can be used for the navigation of planetary rover or other outdoor mobile robots. Originality/value The paper provides a new approach to produce accurate and dense disparity maps of natural terrain, which laid the foundation for its use in outdoor mobile robots navigation.