基于空地影像融合的建筑物结构感知Mesh模型自动构建方法

由于地物遮挡等因素易导致所获取航空影像数据中地物纹理结构信息缺失或不足，现阶段仅利用航空倾斜影像数据自动构建的建筑物三维三角网模型(Mesh)常常会存在模型完整性不足和近地面部分变形较大的问题，因此以航空倾斜影像和地面视角影像为基础，基于空地影像融合的高完整度高精度三维建模已成为当前的一个研究热点。然而为了达到高精度Mesh模型自动构建的要求，就必须解决空地影像的自动匹配问题，实现空地影像数据的一体化高精度处理。本项目系统研究基于空地影像融合的建筑物结构感知Mesh模型自动构建方法。研究内容包括：1）基于物方面元的空地影像自动匹配；2）顾及场景结构特征的多视影像(空地一体化影像)密集匹配；3）具有结构感知功能的高精度Mesh模型自动构建。突破空地影像自动匹配和结构感知高精度建筑物三维模型构建，为自动构建高完整度高精度建筑物Mesh模型提供一条新途径，提升自动化构建的建筑物三维模型的可用性。

Due to the occlusion of ground objects and other factors, some part of the texture structure information of ground objects in the acquired aerial image data is often missing or insufficient. The 3D building mesh model obtained from aerial oblique images by using automatic 3D reconstruction methods may have many problems, such as the 3D mesh model is insufficient completeness (e.g. ground-view holes) and the precision of the 3D model is not high enough (e.g. is with distortion at near-ground parts of the model, the model structure is not prominent). Therefore, based on aerial oblique images and ground-view images, the high-integrity and high-precision 3D reconstruction based on air-ground image fusion is becoming a research hotspot in the field of photogrammetry and computer vision. However, in order to meet the requirements of automatic 3D mesh reconstruction with high accuracy, it is necessary to solve the problem of automatic matching of air-ground images and realize the integrated high-precision processing for air-ground image data. In order to deal with these problems, this project focuses on the automatic 3D mesh reconstruction method with structure sensing functions for buildings based on air-ground image fusion. The research contents include: 1) patch-based automatic image matching method for air-ground images; 2) the self-adaptive patch-based dense matching method of multi-view images (that is, air-ground integrated images) that taking into account the characteristics of the scene structure (e.g., plane, undulating structure, boundary line); 3) automatic high-precision 3D meshing method with structure sensing functions (including the functions of plane simplification and boundary enhancement) from images and point clouds. Our method makes breakthroughs in the automatic image matching of air-ground images and the high-precision 3D building model reconstruction with structure sensing functions, provides a new way for the automatic reconstruction of high-integrity and high-precision building mesh model, and improves the availability of automatic reconstructed 3D building models.