Imagen de Portada

A Synthetic Data Generation Pipeline for Point-Cloud-Based Rebar Segmentation

Guardado en:
Publicado en:ISARC. Proceedings of the International Symposium on Automation and Robotics in Construction vol. 42 (2025), p. 1137-1143
Autor principal: Sun, Tao
Otros Autores: Luo, Yingtong, Shao, Yi
Publicado:
IAARC Publications
Materias:
Rebar
Instance segmentation
Cages
Inspection
Image segmentation
Data collection
Synthetic data
Software
Cameras
Accuracy
Datasets
Computer vision
Experiments
Batch processing
Algorithms
Automation
Annotations
Robotics
Acceso en línea:Citation/Abstract
Full Text - PDF
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
Resumen:Automated rebar cage assembly and quality inspection require reliable rebar recognition. Although rebar segmentation from point clouds has been extensively studied, its generalizability remains limited. One key challenge is the scarcity of real data for training the segmentation models. To address this issue, we propose, for the first time, a pipeline for generating synthetic data for the rebar point cloud instance segmentation task. Using this pipeline, we applied the state-of-the-art Oneformer3d on rebar mesh instance segmentation. The model trained on our synthetic dataset achieved 92.1 mAP in real-world experiments, showing strong synthetic-to-real transfer capability. By eliminating the need for manual data collection and annotation, the proposed method facilitates advancements in automated rebar cage assembly and dimensional quality inspection technologies.
Fuente:Advanced Technologies & Aerospace Database