High-Precision 3D Reconstruction of Historic Buildings Using Multi-Source Data

Historic building documentation requires both complete spatial coverage and reliable geometric detail, but a single surveying technique often cannot meet both requirements in complex heritage scenes. This study proposes a robustness-oriented TLS–UAV point cloud registration and fusion workflow for historic building documentation. The workflow combines feature-based coarse registration with an improved point-to-plane ICP strategy incorporating normal consistency, radiometric correspondence filtering, dynamic distance thresholds, and multi-resolution refinement. The method was evaluated using the Yao’an Lu Junmin Zongguan Fu, a timber–brick courtyard complex in Yunnan, China, under small, medium, and large initial perturbations. Under small and medium perturbations, Point-to-Plane ICP achieved lower RMSE values, while the proposed method produced comparable results. Under large perturbation, the proposed method achieved the highest success rate and the lowest RMSE of 119.0 cm, indicating stronger robustness under challenging initialization. The fused TLS–UAV model achieved checkpoint-based RMSE values of 1.73 cm horizontally and 0.75 cm vertically. Spatial deviation maps showed that residual errors were mainly concentrated around roof edges, eaves, wall corners, and roof–facade transition zones. Cross-scene validation on the Church of Agios Mamas dataset achieved a registration RMSE of 1.2 cm without parameter adjustment. The results show that the proposed workflow is suitable for offline conservation-oriented documentation where registration robustness, model completeness, and component-level geometric interpretation are required.