3D reconstruction and measurement of concrete spalling using near-field Photometric stereo and YOLOv8

Current concrete spalling detection and measurement methods are sparse; despite recent research and commercial offerings using laser scanners, manual measurement is still the industry standard. This paper presents a spalling 3D reconstruction and measurement method. The method uses images illuminated with angled and directional lighting and three neural networks for photometric stereo 3D mesh generation and spalling volume measurement. The proposed method was benchmarked on a laboratory dataset of spalled concrete slabs against a high-resolution laser scanner, yielding an average height error of 0.0 mm and a standard deviation of 1.3 mm. Volume comparisons showed that with manual input, the method achieved a mean absolute percentage error of 22%. Finally, the proposed technique was compared to manual measurements and benchmarked on a spalled concrete structure against a Trimble X12 laser scanner. This research can provide inspectors with increased data interpretability and reduced imaging time for concrete defect mapping.