PSI - Issue 84

Davide Caliò et al. / Procedia Structural Integrity 84 (2026) 513–520

517

dimensional reconstruction. Subsequently, once the point cloud was generated from the properly homogenized thermal data, actual temperature values were assigned point by point to each element of the point cloud. 4. Results 4.1. 3D reconstruction and geometric accuracy Photogrammetric processing was carried out using two distinct pipelines. For the RGB datasets, a conventional workflow based on Structure-from-Motion (SfM) and Multi-View Stereo (MVS) algorithms was adopted (Westoby et al., 2012). The resulting three-dimensional outputs include a high-resolution thermal point cloud of the two inspected piers, consisting of approximately 5 million points, where each point is characterized by thermal information referred to the acquisition time. In addition, two RGB point clouds were generated: one related to the piers and one to the abutment in the direction of Syracuse, consisting of approximately 18 million points and 33 million points, respectively. RGB image acquisition was performed at operational distances ranging between 10 and 35 m, allowing the achievement of an average Ground Sampling Distance (GSD) of approximately 2.5 cm/pixel. The GSD of the thermal point cloud is inevitably higher due to the physical limitations of the thermal sensor, which is characterized by a significantly lower spatial resolution. To mitigate this limitation, thermal acquisitions were conducted at close range and with an image overlap not lower than 70%, achieved through a double acquisition grid with vertical and circular trajectories. This strategy allowed an average thermal GSD of approximately 10 cm/pixel, which was considered adequate for the three-dimensional localization of the main thermal anomalies.

4.2 Structural degradation mapping and thermal anomaly detection (IRT)

The IRT reconstruction enabled a continuous and three-dimensional assessment of the spatial distribution of surface temperatures, overcoming the limitations of individual two-dimensional acquisitions and allowing the identification of persistent thermal anomalies both with vertical development along the pier shafts and at the interfaces between the piers and their foundations.

Figure 2) Comparison between 3D IRT and RGB point-cloud reconstructions of the bridge piers. (a) Thermal (left) and RGB (right) point clouds (b) Close-up of the lower portions of the piers (Inset b), where warm thermal anomalies are associated with potentially bulging concrete cover (c) Close-up of linear cold thermal anomalies along the pier shaft (Inset c)