基于航空点云数据的大范围城区建筑物层次拓扑识别与模型重建

Since 3D building models are important framework data for the digital city, their acquisition issues have drawn numbers of attentions during the past two decades. However，the automatic reconstruction of 3D buildings models still suffers lots of difficulties, such as the scene complexity, the building structure diversity, as well as the difficulties in features extraction and topologies construction..In this project, we attempt to study the issues of topologies construction and model reconstruction towards the complex structure buildings under difficult situations, which including the following points:.Firstly, in allusion to the limitation of current methods in facets features extraction, an deep learning based object-level RANSAC approach is proposed, which is not only of high efficiency and robustness but also adapt to different roof plane type, .Secondly, in allusion to the difficulties in roof topology distinguishing and representation by current methods, such as the omitting and falsely distinguished connections, we propose a hierarchical roof topology-based approach for robust topology reconstruction, which can significantly improve the topology precision and quality..Finally, a model-to-model repairing strategy is designed, which can well improve the eventually reconstruction results..Through this research, a well-organized building reconstruction procedure that can automatically generate the vector models of large area urban buildings from aerial point clouds can be proposed, which can then be adopted the construction of digital cities.

建筑物三维模型作为数字城市中重要的框架数据，其重建工作长久以来受到了广泛而深入的研究。但是由于场景的复杂性、建筑物结构多样性、特征与拓扑识别的困难等原因，目前基于点云数据的建筑物自动重建工作仍然面临较大困难。本项目拟开展针对复杂建筑物的层次拓扑构建以及模型重建方法的研究：首先，针对已有点云分割算法的不足，提出了一种深度学习引导的对象级RANSAC分割方法，在保证高效率与可靠性的同时，能有效识别不同的面片类型；然后针对传统基于拓扑图的拓扑构建方法普遍存在的错检与漏检问题，提出了一种基于层次拓扑树的拓扑构建流程，有效提高了拓扑识别结果的可靠性；最后，实现了一种模型到模型的漏洞修复策略，很好地改善了重建的最终结果。通过本课题的研究，可望提出一整套从大范围建筑物点云出发到矢量化模型生成的自动化重建方案，应用到数字城市的建设中去。

建筑物三维模型作为数字城市中重要的框架数据，其重建工作长久以来受到了广泛而深入的研究。但是由于场景的复杂性、建筑物结构多样性、特征与拓扑识别的困难等原因，目前基于点云数据的建筑物自动重建工作仍然面临较大困难。本项目拟开展针对复杂建筑物的层次化拓扑构建以及模型重建方法的研究:首先，针对已有点云分割算法的不足，提出了一种面向多尺度多形状多目标的混合投票RANSAC分割方法，在保证高效率与可靠性的同时，能有效识别不同的面片类型; 然后针对建筑物基元边界不准确，基元类型识别困难等问题，提出了融合几何与纹理显著性的模型基元边界提取方法;并以此为基础，实现高效准确的实体模型生成。同时，针对建筑物立面的冲击困难，提出了一种基于草图的窗户模型构建方法，自动化地实现窗户类型的识别及模型参数的拟合。通过对ISPRS测试数据集等开源数据的定量及定性实验，对本课题提出的方法进行对比分析，验证了本文方法的适用性与可靠性。本课题的部分方法，已在江苏测绘局等生产单位获得应用，证明了本文方法能服务与大范围城市场景等实景三维模型生产，创造一定的经济价值。