Computer vision framework for UAV-based defect inspection of existing buildings with cement-mortar facades and BIM integration

Abstract

Defect recognition of existing buildings with cement-mortar facades remains challenging due to diverse defect types, inefficient and risky manual inspection, and missing BIM models. This paper presents an automated framework that combines UAV imagery and computer vision techniques for defect recognition and BIM integration. The framework employs RTK-guided UAV flight for image acquisition, OneFormer3D with DBSCAN for envelope segmentation, YOLOv9 for window detection, Mask2Former with SAHI for defect segmentation, and Revit/Dynamo for automated BIM generation. Validation on a 16-story building achieved 96.97% mAP@50 for envelope segmentation, 92.4% mAP@0.5 for window detection, and 65.12% AP50 for defect segmentation, with defects mapped to a queryable BIM model. This framework reduces manual labor and high-altitude risk, providing a digital O&M foundation for existing buildings without BIM. Future work should improve small-crack accuracy, extend the method to other facade materials, and automate flight path planning.