中文摘要：

本文提出一种基于基本形状以及形状间拓扑关系的点云建筑物重建方法。首先,基于每点的法向量以及点间的距离,利用粗分割-细分割的策略对点云场景进行分割,并利用高斯球的性质完成场景中平面点云的识别;然后,利用平面点云的法向量方向、面积等几何特征完成墙面点云的识别与提取;接着,通过＂投影-极值点＂的方法对单个平面进行重建,并根据平面间的空间临近关系分析建筑物平面之间的拓扑关系,根据拓扑关系对初始重建结果进行修正,自动获得建筑物墙面的重建结果。结果表明,该方法能够自动完成场景中建筑物墙面的重建。

英文摘要：

A comprehensive strategy is proposed to reconstruct a building based on the primitive shape and topological relationship.First,the scene point clouds are segmented using the roughdetail segmentation algorithm based on the normals and the distance between points.The planar clusters are extracted from the segments based on the properties of Gaussian sphere.Second,the walls belonging to the building are recognized and extracted based on the properties such as normal directions and areas.Then,each plane is reconstructed by finding the extreme points after projection.The topological relationships between the planes in the scene are analyzed based on the spatial proximity.The initial reconstructed results are refined based on the topological relationships between planes.The experimental results demonstrate that the building walls can be reconstructed by using our method.