Monocular vision-based correction method for the position of robot screw nail

In the process of satellite assembly, there are errors in the installation and positioning of the cabin board and the frame, which leads to the deviation between the position of the screw screwed by the teaching robot and the actual position of the screw hole. In order to solve this problem, a method of robot screw position and orientation correction based on monocular vision is proposed. The Eye-in-Hand structure was used to capture images online by monocular camera. After filtering and denoising the images, the Canny operator was used to extract the feature contour of nail holes. In view of the fracture of the nail hole feature contour, a fracture contour similarity splicing algorithm was proposed, so as to retain the original information of the nail hole more comprehensively. The image was upgraded to the sub-pixel level, and the position of nail holes in the image was extracted with high precision through sub-pixel edge detection. The homologous relation between the actual image and the desired image is calculated to estimate the pose of the robot when taking the desired image. According to the prior relationship between the desired pose when the robot takes a picture and the desired pose when the robot tightens the screws, the actual robot's tightening screw position was corrected. The deviation threshold of the image and the error threshold of the robot kinematics were set, and then the correction times of the position of the screw tightened by the robot were constrained. Finally, the experimental results show that the proposed method can meet the accuracy requirements of robot screw nails in the satellite assembly process.