The International Association for Automation and Robotics in Construction
Publications / CCC 2025 - Zadar, Croatia
BASIC STUDY OF STEEL STRUCTURE WORK AREA RECOGNITION MODEL FOR FIREPROOFING SPRAY ROBOT USING RGB-D IMAGES
Most of the fireproofing spray work on steel structures in construction sites is still carried out manually, resulting concerns about safety and low productivity. Although robotic systems have been introduced to address these issues, existing approaches often lack autonomous work area recognition and require operator intervention. To overcome these limitations, this study proposes a model for identifying steel structures using RGB-D images and generating improved work area masks for fireproofing spray robot. The proposed model applies image segmentation model to detect steel beams and columns, followed by a two-step mask improvement process. Depth-based filtering removes outlier regions based on percentile thresholds, while HSV-based filtering eliminates color inconsistent background pixels. The proposed model improved the mean IoU from 87.67% with raw segmentation model to 90.99% with cleaner object boundaries. Proposed model can accurately recognize the work area and provide this information to the robot, enabling it to perform tasks autonomously.