3D Monitoring and Anomaly Segmentation for Heritage Artifact Maintenance

This thesis focuses on the development and integration of AI-based methods for 3D data segmentation and analysis. Several deep learning architectures are explored, including MeshCNN and U-Net–based encoders for feature extraction and compression. In parallel, a full-stack application was developed to support 3D object visualization, manual annotation, and interactive manipulation, featuring peer-to-peer communication for server-side rendering. Finally, the proposed methods and tools are integrated into the existing software ecosystem provided by Aerariumchain.