A unified framework for 3D building surface reconstruction from fused LiDAR and photogrammetric data

Abstract

Reliable three-dimensional (3D) building surface reconstruction depends on accurate and comprehensive spatial data; however, acquiring such data remains challenging, particularly for point clouds. Data collected using terrestrial laser scanning (TLS) and unmanned aerial vehicle (UAV)-based LiDAR provide detailed geometric information but are often affected by occlusions, limited viewpoints, and sensor constraints, resulting in incomplete datasets. Although data fusion using the iterative closest point (ICP) algorithm and Laplacian-based enhancement can mitigate these limitations, existing research typically treats fusion and surface reconstruction as separate processes, often applying reconstruction directly to fused data without prior quality assessment. This limits the generation of complete and geometrically consistent 3D building models from heterogeneous datasets. To address this gap, this study proposes a unified framework for 3D building surface reconstruction from fused point clouds derived from TLS, UAV LiDAR, and photogrammetric sources. The workflow applies ICP for alignment, Laplacian-based enhancement for gap reduction, and quantitative evaluation of the enhanced point cloud prior to reconstruction using the Marching Cubes algorithm. The results show an average completeness of 86.9%, with the standard deviation improving from 0.173 m to 0.157 m (roof) and from 0.238 m to 0.191 m (wall). The reconstructed model satisfies level of detail 2 (LoD2) requirements. Although certain architectural features, such as windows and pillars, are preserved, they are not yet semantically structured, indicating potential for extension toward higher LoDs (e.g., LoD3). Overall, the framework highlights the importance of pre-reconstruction enhancement and evaluation and demonstrates its potential for supporting more complete and geometrically consistent 3D building modelling.