This study presents an investigation of automatic crack detection at the Victoria Gallery & Museum using 3D laser scanning and point cloud analysis. The workflow comprised three main stages: data collection, processing, and analysis. Terrestrial laser scanning was conducted using a FARO FOCUS 150S to capture dense 3D point clouds of both interior and exterior building sections. Data processing was performed using Scene and CloudCompare software, including scan treatment, segmentation, and alignment, to prepare the datasets for crack detection. Target areas for analysis were identified through prior visual inspection, with segmentation applied to reduce point cloud size and enable efficient Canupo classification. Automatic crack detection was performed using the Canupo method, with training classes carefully created and, in some cases, multiple classification runs required to distinguish cracks from floor, brick, or joint materials accurately effectively within the trained classification framework. For the interior, the first-floor point cloud was reduced from over 13 million points to focused crack segments of approximately 475,239 points, significantly improving processing time and classification performance. Exterior analysis was limited by scaffolding but successfully performed on available sections. The final detected cracks were merged with the original RGB imagery to provide a clear visual representation, demonstrating that the proposed workflow enables effective automated crack identification within the analysed datasets, even under hardware and site constraints. The reported classification results are based on CANUPO performance within the trained datasets and were evaluated through qualitative visual inspection rather than independent quantitative validation. The contribution of this study lies in the development and demonstration of a practical workflow for crack detection in complex heritage environments using TLS and point cloud classification.
Mohammed Al-Shibli (Fri,) studied this question.