Automatic Building Abstraction from Aerial Photogrammetry
Automatic Building Abstraction from Aerial Photogrammetry
复制标题
DOI：
10.5194/isprs-annals-iv-2-w4-243-2017
复制
发表时间：
2017-09
期刊：
影响因子：
--
通讯作者：
A. Ley;R. Hänsch;O. Hellwich
中科院分区：
文献类型：
--
作者：
A. Ley;R. Hänsch;O. Hellwich
摘要。多视图立体视觉已被证明是创建逼真的3D城市模型的一种可行工具。然而，它仍然面临重大挑战，因为当应用于航空图像时，它会产生密集但有噪声且不完整的点云。3D城市建模通常需要与这些点云不同的3D场景表示形式。本文应用一种全自动流程从给定的密集点云生成简化的网格。该网格提供了一定程度的抽象，因为它仅由相对较大的平面和有纹理的表面组成。因此，可以在保持高精度的同时去除噪声、异常值以及杂乱部分。
Abstract. Multi-view stereo has been shown to be a viable tool for the creation of realistic 3D city models. Nevertheless, it still states significant challenges since it results in dense, but noisy and incomplete point clouds when applied to aerial images. 3D city modelling usually requires a different representation of the 3D scene than these point clouds. This paper applies a fully-automatic pipeline to generate a simplified mesh from a given dense point cloud. The mesh provides a certain level of abstraction as it only consists of relatively large planar and textured surfaces. Thus, it is possible to remove noise, outlier, as well as clutter, while maintaining a high level of accuracy.