Abstract In the field of robotic automated welding, the welding of non-standard steel structural components is always a focus of attention. The base of an electric power tower was chosen as the focus for automating the extraction of weld characteristics. A new method combines the YOLO-CE model, an improved 2D image instance segmentation algorithm based on YOLOV8, with 3D point cloud technology. This method uses the YOLO-CE model to accurately extract point cloud data from the target area, followed by applying the maximum likelihood sample consensus (MSAC) algorithm for efficient plane segmentation. Weld lines are located using plane equations, allowing for the initial extraction of the weld point cloud. Precision is improved by developing an optimized evaluation equation that considers the distance between the weld point cloud and the fitted plane and the angle between their normal vectors. This enables detailed classification of the weld point cloud. Based on these classifications, characteristic points of the weld are extracted, and the position of the welding characteristic points is accurately determined by calculating the distance between them and their intersection with three planes. The proposed method’s reliability was verified using a robot for precise measurements, showing a total error of less than 1.5084 mm, indicating high stability and precision. Post-operation inspections confirmed that the welds were filled and free from pores, meeting process requirements. This research offers an efficient and precise solution for the automated welding of non-standard steel structural components, with broad application prospects.