A Single Camera Stereo System for Obstacle Detection

摘要

This paper presents a real-tiem visual obstacle detection system which is novel in both sensors and algorithms. Single-camera stereo vision is realized by using a set of handily placed mirrors, and the optical and electrical identities of binocular image pair are guaranteed. A software gaze transformation method is presented to produce zero disparities for road surface figures, thus facilities the obstacle detection. The proposed non-zero disparity filtering (NZDF) algorithm makes the obstacle detection simple and robust. Problems of feature detection and correspondence are avoided and the system is robust, near real-time, and applicable to natural scene environments. Experimental results are given for outdoor real scene images.