顾及倾斜影像透视变形的Mesh模型构建与优化方法研究

Oblique photography can obtain more ground information, which makes up the weakness of the vertical photography and intuitively provides a realistic 3D model. But it still has some problems: ① the edge features of Mesh model are not prominent enough; ② the building surfaces are local deformed; ③ the Mesh model have holes etc. Therefore, it is still a difficult task that automatically constructs Mesh model based on the oblique images. In this project, a dense matching algorithm is proposed to improve the precision of the depth map for the buildings in the urban area. Then, using the proposed depth map fusion method based on the virtual perspective to remove the point cloud noise, improve the precision of the point cloud and effectively reduce the computational redundancy. Finally, the initial Mesh model constructed by point cloud usually reconstructs an approximate surface, which needs to make use of the image information to compensate for the information loss in the process of Mesh reconstruction. By using the proposed single parameter Mesh optimization algorithm and the regularized term constraint algorithm with feature constraints, a fine Mesh model is constructed to solve the problems of high degree of freedom, non-strict derivation process, weak edge feature and surface deformation in traditional methods. The technology and method studied in this project can be applied to 3D reconstruction of large area and complex scenes. It has great theoretical significance and practical value, and has a wide range of applications in large-scale mapping and urban planning.

倾斜影像弥补了垂直摄影的不足，能够获取更多的地物信息，自动构建完整逼真的三维模型。但是，倾斜影像三维重建还存在一些问题：①棱角特征不够突出；②建筑物表面变形；③Mesh模型中存在孔洞等。因此，基于倾斜影像重建精细的Mesh模型仍然是一个研究难点：本项目提出顾及物方平面的密集匹配算法，针对城市区域地表建筑物，提高深度图的精度；然后，通过虚拟视角深度图融合方法去除点云噪声、提高精度以及有效减少计算冗余；最后，针对点云构建的初始Mesh模型是一个近似表面的问题，需要利用影像信息弥补Mesh重建过程中的信息损失，提出法向单参数法Mesh优化和顾及特征的正则项约束方法，构建精细Mesh模型，解决传统优化方法中自由度高、求导过程不严格、不能保持棱角特征和表面变形等问题。本项目研究的技术与方法适用于大区域、复杂场景的三维重建，具有较大的理论意义和实用价值，在大比例尺测图和城市规划等领域有广泛的应用前景。

倾斜影像的一个重要特点是存在较大的透视变形，给影像匹配带来了一定的困难，从而影响了建模结果。针对倾斜影像三维建模存在的棱角特征不够突出、建筑物表面变形、网格模型中存在孔洞等问题，本项目提出顾及物方平面的密集匹配算法，通过生成透视变形最小的核线影像，最大程度降低了透视变形对密集匹配的影响，针对城市区域地表建筑物，有效提高了深度图的精度；然后，通过深度图融合去除点云噪声，提高点云精度，并增强深度图之间的一致性，本项目通过定义有限数量的虚拟视角，尽可能地减少冗余计算并覆盖全部场景，在虚拟视角上融合深度图并实现三维重建；最后，针对点云构建的初始Mesh模型是一个近似表面的问题，本项目利用影像信息弥补Mesh重建过程中的信息损失，提出法向单参数法Mesh优化和顾及特征的正则项约束方法，有效构建了精细的Mesh模型，解决了传统优化方法中自由度高、求导过程不严格、不能保持棱角特征和表面变形等问题。在本课题的资助下，发表二区SCI论文3篇，申请发明专利3项，本项目研究的技术与方法适用于大区域、复杂场景的三维重建，具有较大的理论意义和实用价值，在大比例尺测图和城市规划等领域有广泛的应用前景。