PSI - Issue 84

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

61

Vegetation classification is carried out through density-based clustering and return-intensity analysis. In multi return LiDAR datasets, the combination of first and last returns helps discriminate between canopy, trunk, and understory elements, further refining the separation between natural and structural classes. A manual editing phase is applied where automated classification remains ambiguous, for example, at pier-embankment interfaces, at transitions between inclined retaining surfaces and natural slopes, or in shadowed zones where return density decreases. The resulting classified point cloud enables the downstream computation of DTMs and DSMs, the extraction of structural geometries, the identification of boundaries, and the support for comparative multi-temporal analyses. After classification, a high-resolution triangulated mesh is produced to provide a continuous and topologically coherent representation of the surveyed environment. Mesh generation typically employs constrained Delaunay triangulation or Poisson-based surface reconstruction to ensure uniformity, robustness to noise, and preservation of sharp features. Structural components, such as pier edges, deck soffit boundaries, and abutment transitions, are maintained through feature-preserving refinement steps that prevent excessive smoothing of corners or abrupt changes in surface curvature. The mesh is validated through comparison with independent checkpoints, extracted from GNSS measurements or ground-derived profiles. These checks quantify planimetric and altimetric deviations, allowing for the assessment of local distortions in regions subject to occlusions or lower return density. Once validated, the mesh provides a reliable geometric basis for engineering applications. BIM-oriented processing can convert structural surfaces into parametric objects, while finite-element workflows benefit from the availability of as-built geometries that capture deviations from idealised design models. The mesh also supports the development of digital-twin components, enabling temporal comparison of successive inspections and integration with structural health monitoring data to detect localised deformation, settlement, or material loss. Compared with photogrammetric meshes, LiDAR-derived surfaces exhibit higher completeness in low-texture or shadowed zones, improved reconstruction of vertical and overhanging elements, and better penetration in vegetated environments. These properties are critical for bridges, where visibility conditions are heterogeneous and significant portions of the structure, such as the soffit, the pier-deck interface are often inaccessible to image-based methods. 4. Case Study and Data Elaboration The methodology was applied to the Viadotto Fiacchignano, a multi-span viaduct located along the SS675 “Umbro Laziale” highway, in Narni (Umbria, Central Italy). The structure spans a deeply incised valley and is surrounded by dense vegetation, steep embankments, and varied terrain conditions. These characteristics make Fiacchignano a suitable benchmark for evaluating airborne LiDAR performance in scenarios where visibility constraints, variable elevations, and the presence of obstacles challenge traditional inspection methods. The structure is made of simply supported prestressed concrete beams. The deck is supported by reinforced concrete piers of variable height, exceeding 30 m in some points, and the deck is positioned above steep valley flanks. The presence of both engineered and natural slopes, along with localised access restrictions beneath the spans, reproduces conditions commonly encountered in the Italian highway network, where the combination of altitude, vegetation, and operational constraints often limits inspection activities. Fiacchignano therefore provides a representative environment for testing the capacity of LiDAR to capture (i) under-deck surfaces, (ii) pier–deck interfaces, (iii) steep and partially vegetated slopes, and (iv) geometric transitions between structural and geotechnical elements. 4.1. UAS-LiDAR Survey A multirotor UAS equipped with a high-frequency LiDAR payload and GNSS–RTK positioning was deployed for the Fiacchignano survey. A pre-flight reconnaissance highlighted several operational constraints. Firstly, the SS675 is a busy arterial road, and no closure of the carriageway was possible during daytime operations, thus restricting direct over-deck or over-road flight paths. Secondly, the valley morphology creates GNSS shadowing conditions beneath the deck, especially near the tallest piers, and required the planning of redundant strips to ensure trajectory stability. Third, the presence of medium to high vegetation along the slopes required the use of multi-return LiDAR acquisition to maximise terrain penetration. The acquisition strategy combined medium-altitude nadir flights for global geometry, longitudinal strips aligned with the deck to stabilise strip adjustment, lateral and oblique passes up