Abstract: Aiming at the shortcomings of many types of steel structure parts, small batches, poor consistency of weld shape and position, and complicated repetitive positioning process of robots, a laser vision-based steel structure weld image processing system is designed. The technology uses CCD industrial cameras and lasers, Collect welding seam images with laser stripes, use median filtering to soften noise, otsu algorithm adaptive threshold segmentation, open operation and morphological processing to remove small connected areas in the image except target pixels, extract laser stripes Finally, the Hough transform is used to fit the center line of the center line to obtain the position of the feature point, and the feasibility of the technology is verified by the bone scaffold test. The test shows that this method can detect the weld feature points quickly and accurately actual requirements, and meet the actual requirements.