Automated LoD-3 Reconstruction Using Oblique UAV Images

Abstract. Urban modeling has increasingly gathered attention for urban resilience simulation studies. However, existing methodologies often fall short in reconstructing detailed building facades necessary for Level of Detail (LoD) 3 modeling. This study presents an automated LoD-3 reconstruction framework using oblique UAV images, which remains underexplored for efficient and scalable LoD-3 reconstruction. The proposed method integrates photogrammetry-based processing with deep learning-based window detection. Our approach consists of extracting roof structures, generating simulated facade images, detecting windows , and mesh intersections. Based on the photogrammetry, cameras are simulated to generate facade images of buildings. The YOLOv5-based window detection is followed using these simulated images. A ray-mesh intersection algorithm is implemented by projecting detected bounding boxes of windows onto the reconstructed LoD-2 model. The final LoD-3 model is exported in CityJSON format for seamless integration into urban simulation applications. Experimental results demonstrate that this approach shows the feasibility of an efficient and scalable solution for large-scale LoD-3 reconstructions.