Abstract: With the substantial increase of railway transportation volume in recent years， railway transportation automation has been scheduled. Nowadays， the efficiency of train shunting and marshalling has been greatly improved. However， the unhooking and decomposition operation of trains still need to be completed manually. In view of short time unhooking operation and the need to accurately identify the target handle and ranging when the robot is synchronized with the carriage， a method of automatic unhooking using binocular stereoscopic vision technology combined with the manipulator is proposed， and the vision part is deeply studied: identifying the target handle through image preprocessing and template matching technology， recovering the pose information between binocular cameras by feature detection and matching algorithm， and correcting the pose information of binocular cameras. In order to overcome the influence of uneven illumination in stereoscopic matching， a stereoscopic matching algorithm based on local fusion is proposed to obtain the disparity map. Finally， triangulation is used to calculate the depth information of the target handle in the train decomposition area to realize three-dimensional reconstruction. This method can measure the physical distance between the handles and the binocular vision system after identifying the position of the handles， so as to provide data basis for the automatic unhooking of the robot.

Key words: automatic unhooking robot; binocular stereoscopic vision; image preprocessing; stereoscopic matching; binocular ranging