Abstract. This paper presents the application of the open-source software Seg4D® for the automated diagnosis of historic constructions using 3D point clouds. Seg4D® integrates machine learning, deep learning as well as several ad-hoc algorithms within the CloudCompare® interface to segment construction systems and detect damage. This software has been tested on the 13th-century Church of San Pedro in Becerril del Carpio, Spain. This building shows prominent structural problems, with relevant out-of-plane deformations. A 3D point cloud containing 33 million points was used as input. The construction systems of this building have been identified by applying the Random Forest algorithm implemented in Seg4D, achieving an average precision of 95%. Subsequent geometric analyses identified out-of-plane displacements, buttress inclinations, and arch deformations. These results demonstrate Seg4D's potential as a holistic tool for semantic segmentation and damage assessment, which facilitates structural diagnosis and conservation in cultural heritage contexts.
Sanz-Honrado et al. (Thu,) studied this question.