The work aims to build a robot visual recognition and handling system based on stereo CCD camera, robot and gripping claws to instruct the robot to achieve automatic grabbing. The OPENCV library functions were used to complete the camera calibration based on the secondary development. Through image acquisition and pre-processing, the algorithms of stereo matching and edge extraction were studied, and the feature contour points of the objects were also extracted to calculate the 3D coordinates of centroid points. The 3D coordinates of the centroid points were converted into robot's joint angle data to provide visual positioning for robotic grasping. Based on VS2012 development platform, visual image processing, robot control and data transmission were integrated into the PC to facilitate human-computer interaction and software integration. The orthogonal experiment indicated that the positioning error in different directions of the space was within the range of 0.5～1.5 mm. The proposed visual guidance scheme can improve the accuracy of robotic motion, and meet the requirements of industrial handling.