A simplification method for large-scale urban point clouds considering diversity of terrain object features

Large-scale, high-precision urban point clouds are extensively utilized across various fields and serve as a fundamental data source for constructing digital twin cities. A critical challenge associated with these data is the substantial volume and high redundancy of original 3D point cloud data, which pose significant difficulties for subsequent applications. Existing 3D point cloud simplification algorithms predominantly focus on simple point cloud objects and often fail to support large-scale data processing while preserving diverse geometric features. In this study, a large-scale urban point cloud simplification method is introduced, which considers the diversity of terrain object features. This method utilizes semantic information to simplify each type of terrain object individually based on its distinct geometric characteristics. Comparative experimental results indicate that compared to baseline algorithms that consider geometric features on the test dataset, the proposed method achieves about 16.5-min processing time for 13-million-point dataset, significantly faster than other algorithms. Additionally, it effectively preserves geometric features, providing improved visual representation, while the geometric similarity metrics demonstrated an average reduction of 80% (Hausdorff distance) and 10% (Chamfer distance). This study presents a novel approach to simplifying large-scale, high-precision urban point clouds and represents a significant advancement in processing and applying massive urban data.