Optimising 3D point cloud semantic segmentation: ML and manual refinement in the UNESCO Saltaire Village

Researchers have demonstrated that combining machine learning (ML) algorithms with manual refinement significantly improves the semantic segmentation of 3D point clouds, particularly for mid-19th-century industrial housing architecture. This approach, tested at the UNESCO World Heritage Site Saltaire Village, addresses challenges posed by contemporary human activity altering the original fabric and spatial composition. The study compares standalone ML and multilayer ML classification methods, providing evidence of difficulties in transitioning from traditional human-led models to HBIM (History Building Information Modeling) in densely altered heritage environments. The findings lay the groundwork for a semi-automated AI-backed scan-to-BIM approach, offering deeper insights into 19th-century architecture and supporting sustainable heritage preservation.

Cotella, V. A., Neagu, C. D., Bavani, S. A., Trichard, M., Horcholle, F., Sangoï, R., Lacalle, C., Jeanvoine, A., Sparrow, T., & Wilson, A. S. (2025). Optimising 3D point cloud semantic segmentation: ML and manual refinement in the UNESCO Saltaire Village. Building Research & Information, 53(7), 846–870. https://doi.org/10.1080/09613218.2025.2521751