Aiming at the problems of classification, recognition and pose estimation of the pipe fitting in robot grasping, the template matching recognition, stereo matching, and threedimensional pose estimation of randomly placed pipe fittings were studied. The internal and external parameters of the left and right cameras were obtained through calibration, and then binocular calibration, correction and handeye calibration were performed. The template matching algorithm was optimized, the edge contours of pipe fittings in template matching were extracted with subpixel accuracy. The appropriate number of image pyramid layers was selected to improve the matching speed and realize the classification and recognition of multiple types of pipe fittings. By using stereo matching, the disparity map of the pipe was extracted,the calibration parameters and the triangulation principle of binocular vision were combined to achieve the pose estimation of the pipe fitting, and multiple sets of pose extraction experiments were carried out. The pipe fitting identification and attitude estimation system was developed by using MFC and Halcon joint programming.The experimental results show that the system can be used to realize realtime and accurate classification and pose estimation of the pipe fittings under random placement, which meets the needs of practical applications in industry.