Point clouds have become the primary medium for documenting and analysing archaeological contexts, driven by the growing availability of high-resolution three-dimensional data. However, the integration and optimisation of multiple point clouds derived from multi-sensor acquisition campaigns remain challenging, particularly with respect to metric consistency, noise management, and the preservation of archaeologically significant micro-morphologies. This study proposes a comprehensive methodological framework for the generation, integration, and enhancement of point clouds obtained through UAV-based photogrammetry and Terrestrial Laser Scanning (TLS). The framework incorporates supervised and weakly supervised Artificial Intelligence modules as adaptive support tools embedded within a coherent geomatic workflow, rather than as full substitutes for established survey methodologies. TLS data are employed as a robust metric reference to improve local reliability assessment, data integration, and targeted denoising, while ensuring traceability, interpretability, and compliance with current regulatory standards. The proposed methodology is validated through its application at the archaeological site of Laurelli (Caselle in Pittari, Southern Italy), a context characterised by complex geometries, heterogeneous surface conditions, and variable stratigraphic features. In the presented case study, quantitative results show that AI-guided UAV–TLS fusion reduces the C2C discrepancy with respect to the adopted TLS metric reference by up to 29% compared with non-adaptive fusion approaches, an improvement in relative geometric consistency rather than an equivalent gain in absolute metric accuracy. A sensitivity analysis across voxel resolutions shows that this benefit is largest at fine resolutions and decreases when both datasets are resampled more coarsely than the UAV point spacing. Furthermore, supervised denoising effectively reduces photogrammetric noise while preserving critical morphological discontinuities that are essential for archaeological interpretation. Overall, the results indicate that the controlled and methodologically grounded integration of AI into established survey practices can significantly enhance the quality, coherence, and usability of three-dimensional archaeological datasets. This, in turn, supports advanced analysis, documentation, and seamless integration into GIS, HBIM, and XR environments. The proposed framework offers a replicable and scalable methodological foundation for future applications in the field of digital cultural heritage.
A Quality-Guided AI Framework for UAV–TLS Point Cloud Integration in Archaeological Contexts
lorusso angelo
;michele pellegrino
;michele scafuro
;maria luigia rizzo
;antonia serritella
2026
Abstract
Point clouds have become the primary medium for documenting and analysing archaeological contexts, driven by the growing availability of high-resolution three-dimensional data. However, the integration and optimisation of multiple point clouds derived from multi-sensor acquisition campaigns remain challenging, particularly with respect to metric consistency, noise management, and the preservation of archaeologically significant micro-morphologies. This study proposes a comprehensive methodological framework for the generation, integration, and enhancement of point clouds obtained through UAV-based photogrammetry and Terrestrial Laser Scanning (TLS). The framework incorporates supervised and weakly supervised Artificial Intelligence modules as adaptive support tools embedded within a coherent geomatic workflow, rather than as full substitutes for established survey methodologies. TLS data are employed as a robust metric reference to improve local reliability assessment, data integration, and targeted denoising, while ensuring traceability, interpretability, and compliance with current regulatory standards. The proposed methodology is validated through its application at the archaeological site of Laurelli (Caselle in Pittari, Southern Italy), a context characterised by complex geometries, heterogeneous surface conditions, and variable stratigraphic features. In the presented case study, quantitative results show that AI-guided UAV–TLS fusion reduces the C2C discrepancy with respect to the adopted TLS metric reference by up to 29% compared with non-adaptive fusion approaches, an improvement in relative geometric consistency rather than an equivalent gain in absolute metric accuracy. A sensitivity analysis across voxel resolutions shows that this benefit is largest at fine resolutions and decreases when both datasets are resampled more coarsely than the UAV point spacing. Furthermore, supervised denoising effectively reduces photogrammetric noise while preserving critical morphological discontinuities that are essential for archaeological interpretation. Overall, the results indicate that the controlled and methodologically grounded integration of AI into established survey practices can significantly enhance the quality, coherence, and usability of three-dimensional archaeological datasets. This, in turn, supports advanced analysis, documentation, and seamless integration into GIS, HBIM, and XR environments. The proposed framework offers a replicable and scalable methodological foundation for future applications in the field of digital cultural heritage.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.


