结合激光点云与影像的LoD3建筑物窗口自动建模

Automatic Modeling of LoD3 Building Windows Combined with Point Cloud and Image

摘要:

结合建筑物地基激光雷达点云与立面单幅影像，解决窗口几何和结构信息描述困难等关键问题。利用深度学习算法进行影像中窗口的识别与提取，并对窗口影像进行线特征提取与拟合得到窗口的结构类型，同时对建筑物点云数据进行网格划分与窗口轮廓提取，得到窗口的几何参数。根据所提结构与几何信息从窗口模型库中调取窗口模板并修改，实现建筑物窗口的自动建模。所提方法的窗口提取精确度可达89.5%，几何精度在5 cm以内。

Abstract:

This paper addresses key issues such as difficulty in describing window geometry and structural parameters by combining building point clouds with facade image. It uses machine learning algorithms to identify and extract windows in the image， and performs line feature extraction and fitting on the window image to obtain the required window structure type. In addition， it conducts grid division and window contour extraction on the building point cloud to obtain the required geometric parameters. Moreover， based on the extracted structure type and geometric parameters， it retrieves the model template from the window model library and modified to achieve efficient modeling of building windows. The windows extraction accuracy of the proposed method can reach 89.5%， and the geometric accuracy is within 5 cm.