3D modeling of urban buildings using improved DBSCAN algorithm and grid partitioning

Abstract

Large-scale 3D point cloud modeling plays a key role in building recognition and the construction of digital cities. However, existing methods generally suffer from low computational efficiency, insufficient clustering accuracy, and unstable modeling results. Therefore, this study proposes a 3D building modeling approach using Density-Based Spatial Clustering of Application with Noise. The method introduces geodesic distance and neighborhood search strategies, combined with a hierarchical coordination grid partitioning mechanism that considers boundary curvature features, to improve modeling accuracy and processing efficiency. Experimental results show that the model achieves an area under the curve, harmonic mean of precision and recall, standard mutual information, Jaccard index, and adjusted Rand index of 0.984, 0.957, 0.973, 0.9754, and 0.978, respectively, all significantly higher than those of the comparison models. At the same time, the model’s point cloud restoration accuracy, as measured by the root mean square error and mean absolute percentage error, is 4.9% and 6.1%, both of which are significantly lower than those of the three comparison models. These results indicate that the proposed 3D modeling model demonstrates strong advantages in clustering performance, spatial structure restoration ability, and operational efficiency. It shows great scalability and practical value, providing solid technical support for urban digital construction and spatial information reconstruction.