Validation of terrestrial laser scanning and artificial intelligence for measuring deformations of cultural heritage structures

Abstract The paper discusses the issue of using artificial neural networks and point clouds for calculating displacements of cultural heritage structures. The model trained on laboratory dataset was able to determine displacements of the building facade, with the relative accuracy of 3% of the simulated values. The success rate of this model was equal to 85%. The deformations that were derived from digital surface models generated from point clouds, had the relative accuracy of 7% and the values determined by image based close-range photogrammetry methods - 35%. A major novelty is the use of neural networks to determine deformations based on sub-models generated from the point cloud and the unique, supervised-trained, high accuracy predictive model. Practical significance is associated with creating an end-to-end solution that performs detection and estimates the value of the deformation automatically which is a major advantage over other methods.