Stereovision based obstacle detection approach for mobile robot navigation

Abstract

An obstacle detection approach based on stereo vision is proposed for mobile robot navigation. The aim of the approach is to detect a real environment to allow a mobile robot to find a safe path even in complex scenarios. Our novelty is represented in terms of the two-stage perception structure. The detection stage infers the relationship between obstacles and a ground, and determines the region of interest. Based on this, the confirmation stage focuses on characterizing the contours and positions of obstacles and getting rid of most artefacts. The approach synthesizes the statistical information of projections and depth discontinuities in the region of interest to characterize obstacles. The detection-confirmation structure is robust to various lighting conditions, effective to negative obstacles, free to specular reflection effect, and real-time response to the interference from dynamic obstacles. Experimental results verified the effectiveness, reliability and real-time performance of the proposed approach.