融合机载与车载点云的建筑物群快速三维重建方法研究

高质量激光扫描点云数据是当前重构高逼真度建筑物三维模型的一种重要数据源，而如何高精度、全自动、逼真地重构建筑物三维模型是智慧城市建设亟需解决的关键问题。单一数据源存在分辨率与数据覆盖的局限性，采用多源点云数据融合技术是更真实、更科学地实现建筑物重建的重要途径。本课题针对目前建筑物三维重建的数据不完备性、特征提取与重建方法复杂、智能化程度低、鲁棒性差等不足，提出了融合机载和车载点云实现快速三维重建的理论与方法。主要研究内容包括基于线特征的多源点云精确配准方法；基于体元深度学习机制的目标分割与分类方法；基于多视角点云特征图与直线半平面结构的三维直线提取算法以及几何与语义特征相结合的建筑物群三维重建方法。本课题将深层特征感知的深度学习理论应用于高分辨率的融合点云数据解译，将为大场景城区建筑物群的精细三维重建提供新的理论依据和技术支撑。

High-quality laser scanning point cloud data nowadays are an important data source for reconstructing highly detailed 3D building models. Highly accurat, fully automatic, and highly detailed 3D building models are urgently required by today's smart cities. Single data source has limitations in resolution and coverage. Therefore, multi-source point cloud data provide a new means for rapid, accurate, and robust building reconstruction. This project aims at solving problems about data incompleteness, complexity in feature extraction and 3D recontruction methods, low level in intelligence and robustness. The focus of the project is placed on the development of new theories and methods for rapid 3D building reconstruction by fusing airborne and vehicle-borne laser scanning point cloud data. The main research tasks include accurate methods for multi-source point cloud registration based on line features, methods for voxel-based object segmentation and classification using deep learning mechanism, algorithms for 3D line extraction based on multi-view point cloud feature graphs and line-segment-half-plane (LSHP) structures, and methods for 3D building reconstrution through integration of geometric and semantic properties. The project applies the deep learning theory about deep feature sensing to the interpretation of high-resolution fused point cloud data. It is anticipated that the outcomes of the project will provide new theories and technical supports to 3D detailed reconstruction of urban buildings.