In order to meet the need of identifying obstacles and navigating for the Coal Mine Detection Robot which is used to rescue life and detect environment from coal mine disaster, binocular stereo vision is researched and 3D model of objects around the robot is reconstructed by means of two cameras of visual system built in the robot. The two cameras are calibrated and two projection matrices of them are obtained. Then, two images of the same scene are obtained by the two cameras. The matching points of two-dimensional coordinate are got through Harris corner extraction and stereo matching. According to the principle of binocular vision, equations are obtained and solved by least square method, which can calculated the discrete points of 3D coordinate.