Bringing machine learning to archaeology for parametric modeling in BIM

In contemporary archeological survey and project execution, extensive data acquisition techniques are widely employed to capture diverse morphologies of heritage sites. However, 3D digitization and building information modeling (BIM) still demand significant effort, especially in Scan-to-BIM procedures. To address this, artificial intelligence (AI) offers a powerful array of segmentation and classification algorithms capable of automatically identifying and characterizing objects and surfaces. This study investigates the application of machine learning (ML) in the characterization of archeological forms through three critical processes integral to archeological methodology: i) filtering vegetation from the landscape point cloud; ii) semantically segmenting orthostat units within a dolmen structure using the Brodu and Lague morphological segmentation algorithm; and iii) predicting the convex hull of the intrados based on the dolmen’s stone morphology for integration into a BIM model. The research focuses specifically on Dolmen Number 5, located within the Pozuelo complex in Zalamea la Real, Spain. A framework was developed to streamline Scan-to-BIM processes for modeling landscapes, archeological sites, ground surfaces, corridor slabs, and funerary chamber closures. The study evaluates the efficacy of ML algorithms through standard accuracy metrics, with the multiscale algorithm tested across training, testing, and validation phases in a binary classification context. Additionally, a novel ML-based prediction model was proposed to identify the optimal morphological pattern for vertical orthostats. Key findings indicate that a cloth resolution (CR) value of 0.5 significantly enhances the performance of cloth simulation filtering (CSF). The optimal parameter scale for semantically labeling dolmen orthostats using the multiscale algorithm was also established. Moreover, ML models predicted four distinct morphological structures, critical for identifying the convex hull of vertical slabs within funerary chambers, contributing to enhanced BIM modeling.