Monitoring geometric deformations of dam structures has traditionally relied on discrete, pointwise measurements acquired using total stations and predefined signalized targets. While this approach provides high precision at selected locations, it remains spatially sparse and limited in its ability to capture surface-wide deformations. In contrast, dense point clouds obtained from terrestrial laser scanning (TLS) or unmanned aerial vehicle (UAV)-based laser scanners provide high-resolution surface coverage, enabling comprehensive deformation analysis. However, they lack the ability to consistently observe identical surface points across different epochs, which complicates temporal comparisons.

To address this limitation, we propose a feature-based registration workflow tailored for multi-epoch point cloud analysis. Distinct and temporally stable 3D surface features are extracted using established handcrafted keypoint detectors and local geometric descriptors. These features are matched across epochs to establish repeatable point correspondences. Based on the resulting correspondences, a global affine transformation is estimated via a least-squares adjustment to quantify spatially distributed deformations on the dam surface. This approach enables geometrically interpretable deformation models while avoiding the limitations of dense surface matching techniques.

The method is evaluated using real-world data from two different dams, acquired through point clouds from both TLS and UAV platforms. The resulting deformation estimates are compared against those obtained using the Iterative Closest Point (ICP) algorithm, which serves as a widely used reference method.

Authors:
Annika Tobies, Eike Koller, Ansgar Dreier, André Cornelißen, Lasse Klingbeil and Heiner Kuhlmann