| Title |
Development of an RGB-D Camera-based Work Area Recognition Technology for Fireproofing Spray Robot |
| Authors |
Sangmin Lee ; Kangmin Bae ; Minseung Cha ; Sebeen Yoon ; Taehoon Kim |
| DOI |
https://dx.doi.org/10.6106/KJCEM.2026.27.5.135 |
| Keywords |
Fireproofing Spray Robot; RGB-D Camera; Steel Member Recognition; Work Area Classification |
| Abstract |
Fireproofing spray on steel structures is essential for securing building fire resistance. However, conventional fireproofing spray is often performed manually at elevated locations and under dusty conditions, posing safety risks and causing quality variations depending on spray conditions and worker skill. Existing fireproofing spray robot studies have mainly relied on predefined work maps or structural design data, or have recognized steel members only at the object level. These approaches have limitations in providing work surface information required for robotic spray path generation. This study proposes an RGB-D camera-based work area recognition method for autonomous fireproofing spray robots. The proposed method first applies YOLOv11-based instance segmentation to RGB images to recognize the steel member and generate an initial mask. The mask is then refined through depth-based refinement and HSV-based refinement to extract the work area requiring spraying. The extracted work area is generated as a 3D point cloud using the depth map and camera intrinsic parameters. Finally, the work area is classified into five work surfaces using plane estimation with RANSAC and clustering with DBSCAN. The method was validated using 30 pairs of RGB images and depth maps collected from an MDF-based H-shaped steel beam mock-up using an Intel RealSense D455 camera. The results showed a mean IoU of 0.9464 for steel member recognition and a macro F1-score of 0.8558 for work surface classification. This study provides work area information classified by individual surfaces for spray path generation and construction quality management. |