Method for Initial Point Positioning of V-bevel Welding Seam Based on Binocular Vision

Abstract

Aiming at the positioning problem of welding the initial point in the process of welding medium-thick plate V- bevel, a method for positioning based on binocular vision is proposed in this paper. The binocular cameras are arranged in an eye-to-hand manner to capture images included the initial point of the workpiece weld, using YOLOv5 neural network to localize the area included the welding initial point. Then the image of positioned area is preprocessed, using the straight line detection method to extract the workpiece edge on the image captured by the binocular camera to find the starting point of the tack weld. Then get the data of stereo calibration the binocular camera and eye-to-hand calibration between robot and camera to determine the relationship between the picture pixel coordinate system and the robot base coordinate system. Finally, the coordinates of the starting point of the welding seam in the robot base coordinate system are calculated using the calibration results and the pixel coordinates on the starting point picture of the welding seam. After comparison with the coordinates demonstrated by the robot, the precision of the guidance method proved to be able to meet the requirements for the use of subsequent seam tracking welds.