A suitable visual aid system is needed to judge the position and distance of the robot arm or claw in the high-altitude live working robot"s fine work on the target object. This paper presents a binocular stereo vision based target recognition and localization method for live robot. Among them, the traditional Census algorithm is improved by replacing the center pixel with the window mean pixel and the fixed window with the adaptive window, and weighted fusion with SAD algorithm is obtained to obtain an improved Census-SAD stereo matching fusion algorithm. The algorithm is used for pixel matching and parallax map is obtained. Train the bolt data set of transmission lines under the YOLO framework to obtain the bolt depth distance and accurate identification. The experimental results of bolt positioning show that the proposed method can achieve a positioning accuracy of 1.2% in close range positioning,the minimum relative error was 0.5%, which can provide real-time and accurate environment perception information for live robot, and improve the operating stability of live robot.