A stereo vision-based method was proposed to measure the relative poses (position and attitude) between the two non-cooperative spacecrafts to solve the problem of the non-cooperative target recognition for the on-orbital maintenance and space debris removal. A median filter was used to smooth the original images by removing the disturbance of orbital background and other noises.The edges in the smoothed images were detected using the Canny edge detector to obtain the corresponding binary images.Then,the straight lines of the object to be recognized were extracted from the binary images through Hough transform,and the intersection points among these lines were determined.Finally,the feature points in the left and right images were reconstructed in 3D to obtain the coordinates of the object points in the world frame.Based on these points,the target frame was defined and the pose of the target with respect to the world frame was determined.Simulation results indicate that the position errors are lower than 40 mm in the farther distance (>2.5 m) and 10 mm in the nearer distance (<2.0 m),and the orientation errors are less than 2°,which satisfies the requirements of pose measurements during the tracking,approaching and flying round the non-cooperative target.