An Automatic Building Models' Parametrer Reconstruction Method from Point Clouds.

Compared with point clouds, three-dimensional (3D) models can provide geographic information features for more efficient processing, retrieval, exchange and visualization. However, the construction of 3D models, especially large-scale outdoor scenes, requires expensive time and human resources. In contrast to traditional methods, this paper proposes a semantic segmentation network with hierarchical understanding and employs predefined components to reconstruct the building model, in which all point clouds of an object are considered at the same time. Experiments show that our method applied to the dataset has an accuracy rate of 89% for the original classification and a mean point-to-surface distance reconstruction quality of 0.06 m is achieved. This study also demonstrates the effectiveness of the proposed method and its potential to generate 3D models from large-scale urban point clouds.