Abstract: Considering the environmental characters of welding in unstructured space, via vision technology, a twosteps template matching method for workpiece recognition based on circle projection and Zernike moments was presented. An algorithm based on Hough line detection was designed to extract special lines which were used to calculate initial seam position with subpixel precision. Experiment was carried out by mobile welding robot, and series of templates were acquired by snapping workpiece images with different rotation angles. Circle projection results and Zernike moments were calculated, and they could effectively represent features of templates for typical workpiece, meanwhile errors due to rotation could be eliminated. These algorithms were proved to perform satisfied robustness of illumination and camera view. Subpixel location of initial seam position can satisfy the need for visual calculation.

Key words: autonomous welding, mobile welding robot, vision technology, workpiece regcognition