Leveraging Historical Archives with Artificial Intelligence algorithms for the Digital Reconstruction of Architectural Heritage

Abstract

The preservation and digital reconstruction of architectural heritage remain a significant challenge, particularly when historical records are limited to single images or incomplete datasets. This study explores the potential of AI-driven methodologies for 3D reconstruction capable of generating detailed 3D meshes from a single image. Applying this approach to historical archives data, the model was tested on a selection of case studies in South Tyrol. The results highlight the crucial role of image resolution, detail clarity, and architectural symmetry in determining the accuracy of the reconstruction. Nevertheless, the ability to generate new views from a single image serves as a promising starting point for further model refinement. While challenges remain, particularly with irregular structures and low-quality historical images, this research demonstrates how AI can enhance the study, preservation, and public engagement of architectural heritage. The findings contribute to the ongoing discourse on digital heritage, suggesting avenues for improving AI-driven reconstructions through more diverse training datasets and hybrid methodologies. These technologies provide a framework for transforming static archival records into dynamic, interactive 3D models, offering new opportunities for research, preservation, and public engagement of cultural heritage.

DOI: https://doi.org/10.20365/disegnarecon.34.2025.28