PSI - Issue 84

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

60

integrates raw GNSS observables with inertial measurements to reconstruct a high-accuracy trajectory even in the absence of continuous RTK coverage. PPK adjustment also enhances the internal consistency of overlapping LiDAR strips, which is crucial for subsequent strip alignment. Additional geometric constraints, such as ground control points, redundant scale bars, or reference markers, may be introduced when feasible to further stabilise the solution. Although LiDAR surveys typically rely less on ground control than photogrammetric workflows, these redundant constraints improve reliability in areas where GNSS reception is degraded. By combining carefully selected acquisition parameters, calibrated sensor geometry, and GNSS/IMU data fusion, the methodology ensures that the airborne LiDAR system produces dense, coherent, and metrically reliable point clouds suitable for detailed geometric reconstruction of bridge infrastructures. Survey flights follow the planned trajectories to ensure complete coverage of the bridge and its surroundings. The mission typically combines nadir passes for overall geometry, longitudinal LiDAR strips to enhance internal consistency, lateral or oblique profiles for vertical elements, and low-altitude manual passes to capture under-deck surfaces. This multi-viewpoint strategy reduces occlusions and improves the completeness of the reconstructed geometry. During acquisition, the flight crew monitors GNSS/IMU integrity, point-density uniformity, trajectory continuity, and coverage adequacy. Overlapping flight lines and redundant viewpoints support robust strip adjustment and reduce sensitivity to temporary GNSS degradation. This configuration provides the conditions needed to generate a coherent, high-density LiDAR dataset suitable for subsequent processing. 3. LiDAR Data Processing 3.1. Trajectory Reconstruction and Strip Adjustment The processing begins with trajectory reconstruction, obtained by integrating raw GNSS observations with IMU data to estimate the time-dependent position and orientation of the LiDAR sensor. GNSS data are used to constrain the absolute georeferencing of the flight trajectory, while the IMU provides high-frequency attitude information able to capture rapid variations induced by wind, platform motion, or dynamic corrections. The fusion of these datasets balances the strengths and weaknesses of each sensor: GNSS provides absolute accuracy at low sampling rates, whereas the IMU ensures continuity at high frequency but is subject to drift. A Kalman filter or equivalent smoothing algorithm is typically applied to produce a unified trajectory that is both stable and sufficiently detailed for high resolution structural modelling. Following the reconstruction, strip adjustment is performed to remove residual misalignments between overlapping flight lines. These misalignments may result from GNSS signal degradation, IMU drift, or imperfect boresight calibration. The adjustment process minimises discrepancies in elevation, roll, pitch, and yaw between strips by solving a least-squares optimisation problem that equalises overlapping surfaces. Particular attention is paid to vertical elements, such as pier shafts and deck edges, because small angular deviations may generate visible artefacts in the final point cloud. A correctly adjusted dataset shows uniform point spacing, consistent elevation levels across strips, and the absence of discontinuities on sharp features. 3.2. Point Cloud Classification, Mesh Generation and Digital Modelling The georeferenced point cloud is then subjected to a classification workflow designed to separate structural and natural components. Initially, ground-point identification is performed using height-based filtering and progressive morphological operations, which remove vegetation and isolate the terrain surface. This enables a reliable generation of Digital Terrain Models (DTMs), especially in densely vegetated or partially inaccessible areas beneath the bridge spans. Subsequently, feature-based algorithms relying on surface normals, curvature, and local planarity metrics are used to identify structural elements such as deck slabs, diaphragms, parapets, and pier walls. These elements generally exhibit low roughness and well-defined geometric continuity, which facilitate their distinction from natural features. 2.3. Multi-View LiDAR Acquisition