Abstract: At present, the welding production of prefabricated concrete wind turbine tower molds primarily relies on manual gas-shielded welding. This method faces problems such as low production efficiency and inconsistent product quality. The introduction of robot welding assisted by binocular vision can significantly enhance mold production efficiency and reduce labor intensity. In this study, binocular vision combined with laser-structured light assistance was employed to carry out three-dimensional reconstruction of the weld seams on wind turbine tower molds. Factors affecting the extraction of weld seam point cloud for wind turbine tower mold weld recognition and the extraction of base plate weld seam point cloud were studied. To support the extraction of weld point clouds, the TZ-Dust3r network was developed. Using the TZ-Dust3r network, a 3D reconstruction of the mold and its surrounding environment were completed, thereby establishing a data foundation for subsequent point cloud extraction. By utilizing the geometric characteristics of the mold point cloud, the weld point cloud extraction was accomplished through a projection method integrated with image processing techniques. The binocular vision-based weld seam recognition technology effectively identifies weld seams in irregular wind turbine tower molds, improves the accuracy of weld point clouds, and enables precise seam tracking. It provides crucial technical support for follow-up automated welding processes.