Towards Fully Automated 3D Reconstruction Using UAV LiDAR and Deep Learning
Calvin Wijaya (*), Ruli Andaru, Harintaka, Catur Aries Rokhmana

Department of Geodetic Engineering, Faculty of Engineering, Universitas Gadjah Mada, Jl. Grafika Bulaksumur No.2, Sleman, Indonesia
*calvin.wijaya[at]ugm.ac.id

Abstract

Digital twin technology has emerged as a significant research trend in recent years, offering enhanced capabilities by integrating traditional 2D geospatial data with 3D models and real-time sensor data. This integration enables a more detailed and dynamic representation of physical environments, supporting better-informed decision-making processes. A critical component in the development of a digital twin is the reconstruction of accurate and detailed 3D models. However, this process remains a major challenge due to its complexity and time-consuming nature-from data acquisition to model generation. This research presents and evaluates an automated workflow for 3D model reconstruction, starting from data acquisition using Unmanned Aerial Vehicle (UAV)-based LiDAR systems. The study focuses on the Universitas Gadjah Mada campus area as a case study. The input data consists of UAV-acquired point clouds and high-resolution aerial imagery. We employ a combination of deep learning techniques and geometric processing to preprocess the point cloud data. Specifically, the Dynamic Graph Convolutional Neural Network (DGCNN) is used to classify point clouds into semantic categories such as ground, buildings, and vegetation. Using the classified point cloud data and extracted building outlines, the workflow reconstructs 3D building models automatically. The Geoflow algorithm is applied to generate the final 3D campus model in Level of Detail (LOD) 2.2, encoded in the CityJSON format. This format adheres to Open Geospatial Consortium (OGC) standards, ensuring data interoperability and usability in smart city and digital twin applications. The results demonstrate a streamlined and efficient approach for producing semantically rich 3D city models with minimal manual intervention.

Keywords: 3D Reconstruction, 3D City, Digital Twin, LiDAR, Sustainable Cities

Topic: Topic C: Emerging Technologies in Remote Sensing