Welding seam identification is an important step to realize automatic welding. In view of the lack of automatic identification methods in the existing weld recognition algorithm, an algorithm of weld recognition based on laser vision guidance for various types of welds is proposed. The point cloud image of the weld seam is collected using a laser vision system, and invalid point clouds are removed through point cloud preprocessing. The point cloud segmentation and fitting algorithm is improved to quickly segment the feature areas of the weld seam. And the straight-line models are obtained through segmentation and fitting to autonomously classify the weld seam joints, achieving direct extraction of weld seam feature points. Through testing the extraction of feature points for different types of weld seams, the results show that the algorithm in this paper has an error of less than 0.7 mm in identifying feature points for butt joints, corner joints, T joints, and lap joint, and the identification time of the double characteristic line weld is less than 0.63 s, which not only meets the demand of welding production, but also solves the problem of multi-type welding seam identification and improves the degree of welding intelligence.