Abstract. The accurate reconstruction of 3D building models is essential for urban planning and smart city applications. This study introduces an automated workflow integrating Unmanned Aerial Vehicle (UAV)-derived orthophotos and point clouds to enhance reconstruction accuracy. A deep learning-based tree segmentation model filters non-building objects, while the Cloth Simulation Filter (CSF) separates ground and non-ground points. Clustering techniques isolate building structures, followed by Mobile Line Segment Detector (LSD)-based roof edge detection and refinement. The extracted roof edges are then combined with height attributes to generate 3D bounding boxes. Experiments on UAV data from Suseo, South Korea, show that this approach reconstructs detailed and realistic 3D models, achieving high precision and recall measures. By integrating deep learning, clustering, and geometric analysis, this study provides a scalable and efficient solution for urban modeling.
Bui et al. (Mon,) studied this question.