RT journal article
T1 A Neural-Network-Based Cost-Effective Method for Initial Weld Point Extraction from 2D Images
A1 López Fuster, Miguel Ángel
A1 Morgado Estévez, Arturo
A1 Díaz Cano, Ignacio
A1 Badesa Clemente, Francisco Javier
A2 Ingeniería en AutomáticaElectrónica, Arquitectura y Redes de Computadores
A2 Ingeniería Informática
K1 initial weld point
K1 robotic welding
K1 object detection
K1 robotics
K1 computer vision
K1 point cloud
K1 shipbuilding
K1 intelligent welding
K1 YOLO
AB This paper presents a novel approach for extracting 3D weld point information using atwo-stage deep learning pipeline based on readily available 2D RGB cameras. Our method utilizesYOLOv8s for object detection, specifically targeting vertices, followed by semantic segmentation forprecise pixel localization. This pipeline addresses the challenges posed by low-contrast images andcomplex geometries, significantly reducing costs compared with traditional 3D-based solutions. Wedemonstrated the effectiveness of our approach through a comparison with a 3D-point-cloud-basedmethod, showcasing the potential for improved speed and efficiency. This research advances thefield of automated welding by providing a cost-effective and versatile solution for extracting keyinformation from 2D images.
PB MDPI
SN 2075-1702
YR 2024
FD 2024
LK http://hdl.handle.net/10498/35749
UL http://hdl.handle.net/10498/35749
LA eng
DS Repositorio Institucional de la Universidad de Cádiz
RD 10-may-2026