Abstract: Urban architectural entities are one of the important elements of basic geographic entity production in the construction of 3D real scene China. For the 3D block models of buildings, the existing modeling algorithms and software based on point cloud data generally suffer from low automation, insufficient refinement, and limited scale production,et al. In this paper, we propose an automated 3D reconstruction method for urban building entities. Firstly, the point cloud of independent buildings is identified using the point cloud segmentation instance algorithm, and the bottom surface contour information of the building is automatically extracted and optimized. Then, the segmented point cloud data is managed as a compact structure, and each top surface component point set is obtained through Euclidean clustering and mapped and fitted into polygons. At the same time, the bottom surface contour is introduced as a hard constraint, and the post-processing refinement and merging of the fitted polygons. Finally, the automated 3D reconstruction software for urban building entities is developed and realized, which integrates the capabilities of large-scale modeling, editing, visualization and quality control. The above key technologies and methods have been verified in three 3D real scene construction projects in Ningbo, Nanchang and Chuxiong. The validation results show that it supports the rapid reconstruction of 3D models of building blocks above 10 000 levels, and the modeling speed is better than 16 buildings per second, which can provide strong support for the large-scale production and construction of the building entities of the 3D real scene construction.

Key words: 3D real scene, urban building entities, 3D reconstruction, automation, optimization of top surface components