PSI - Issue 84

Antonio Di Pietro et al. / Procedia Structural Integrity 84 (2026) 57–64

62

to ±45° to illuminate pier shafts and abutments, and a low-altitude manual pass (12–18 m) beneath the deck. This last pass was essential because the under-deck region cannot be adequately observed from nadir or oblique viewpoints due to the height of the structure and the absence of safe ground-based positions for TLS acquisition. A preliminary photogrammetric DSM was also generated to support trajectory planning and identify the most occluded areas. 4.2. Data Processing and Point Cloud Classification Raw data were processed through GNSS–IMU fusion and strip adjustment. The steep valley geometry and variable GNSS visibility resulted in minor inconsistencies between some overlapping strips, which were effectively corrected during the adjustment step. The final point cloud comprised of approximately 160 million points, with densities ranging from 300 to 360 pts/m² along the deck and exceeding 450 pts/m² in proximity to the tallest piers and under deck regions. Classification was particularly relevant at the viaduct site due to the heterogeneity of the terrain. The combination of return-number analysis and local surface metrics allowed discrimination between structural surfaces, bare earth, and multi-layer vegetation along the valley slopes. LiDAR performed effectively in extracting ground points beneath oak and shrub canopies that would traditionally create significant occlusions in photogrammetric surveys. The classification phase also allowed isolation of structural details, including the pier cap, parapets, diaphragms, and slope-retaining structures adjacent to the abutments. Fig. 2 summarises both the classified dataset and the flight trajectories, providing a visual reference for the internal consistency of the acquisition. Ground-point extraction under partial coverage of mixed deciduous vegetation achieved a completeness exceeding 85%, confirming the effectiveness of multi-return LiDAR in vegetated valley environments.

Fig. 2. Fiacchignano Bridge DJI Terra Point Cloud Classification, the green lines are the representation of the UAS track flown during the operation, always under RTK signal cover.

4.3. Geometric Reconstruction and Model Derivation The triangulated mesh derived from the classified dataset provided a continuous reconstruction of all key structural components. Independent RTK checkpoints placed on accessible areas yielded planimetric and vertical RMSE values of approximately 2 cm, consistent with expected performance for airborne RTK-supported LiDAR in complex terrain. In this site, the geometry of the tallest piers and the under-deck region would be difficult to document with conventional methods due to height, slope inclination, and vegetation presence. The LiDAR dataset captured the deck soffit, pier shafts, and deck edges without the discontinuities typically produced by photogrammetric approaches in low-texture, shaded, or occluded areas. The mesh allowed the extraction of alignment profiles, the identification of local surface irregularities, and the generation of terrain models that include both engineered embankments and natural slopes. Fig. 3 illustrates the head of pier 5 as reconstructed from the classified LiDAR point cloud. This level of geometric detail is relevant for structural interpretation, especially for the analysis of the bearing areas.