Impact of Point Density Variation in Aerial Photogrammetric Point Clouds on Feature Extraction for the development of Cultural Heritage Digital Twin

Abstract. The development of Digital Twins for cultural heritage applications depends highly on accurate and detailed 3D representations of historical structures. Aerial photogrammetry has emerged as a popular remote sensing technique for capturing such Cultural Heritage (CH) structures, primarily due to its high-resolution, detailed outputs and cost-effectiveness. However, the quality of the derived photogrammetric point clouds, particularly point density, significantly influences the efficiency and accuracy of downstream procedures such as feature extraction to be used for different applications. This research work investigates how variations in point density affect the detection and segmentation of windows and rooftops, which are two key architectural features for the development of Energy Digital Twins (EDT) for the analysis of energy consumption patterns of the CH buildings. Using aerial photogrammetric datasets, we generated point clouds using the photogrammetric images and their accurate image orientations with a bundle adjustment processing. After that, we tested the point cloud at different densities (ranging from original resolution to 1/16th of the original point density) by controlled uniform down-sampling of an original high-density cloud. We evaluated an automated deep learning-based method segmentation of the windows and rooftop features from the point cloud datasets. The investigation indicates a strong correlation between point density and feature extraction accuracy, with a clear decline in detection performance if the subsampling goes below 1/8th of the original point density, which is around 20 points/m². Rooftop features exhibited greater resilience to reduced density compared to window features and were still detected even with down-sampled point clouds. The research work proposes a density-aware workflow for CH Digital Twin development and emphasises the need for strategic planning in aerial data acquisition for heritage documentation for the optimisation of aerial photogrammetric data practices and to enhance the reliability of Digital Twins in cultural heritage conservation.