实景三维建模兼具直观、可量、可算、信息丰富等特点，是智慧地球建设过程中的 重要应用技术。倾斜摄影三维重建技术因其具有覆盖场景范围广、生成模型效率高、付 出成本代价低等特点，是实景三维模型生产的主要手段之一。但由于倾斜影像获取立面 信息有限，多视影像匹配时缺乏立面特征，生成点云较少，导致模型立面结构不完整， 存在立面凹陷或凸起等现象。为了提高三维模型的真实感以及满足多种空间分析的需求， 需要通过修复手段改善三维模型的可视化效果。已有的修复方法分为半自动与人工修复 两种方式，半自动修复方式中大量噪声的存在使得三维模型房屋边线歪曲，难以完整的 提取并修复建筑物立面；人工交互修复方式成本高且效率低，并不适用于大场景模型。 基于以上背景，本文针对纹理弱、噪声明显的建筑物立面修复问题提出一种智能提 取弱纹理立面及非弱纹理立面的方法，并对三维建筑物立面进行修复，改善三维模型可 视化效果，主要研究内容如下： (1)针对三维模型中弱纹理立面凹陷、不平整问题，提出智能提取弱纹理立面的方法 并进行踏平式修复。基于倾斜影像多片信息，结合超像素分割方法提取影像中的弱纹理 像素，将反投影与遮挡判断相结合，获取像素与三维 Mesh 模型中三角面之间的多对一 关系，即 faceMap，从而建立二维弱纹理像素与三维弱纹理立面之间的联系，再通过平 面投影方式对弱纹理立面进行修复。利用本文方法修复后，立面平整度明显提升，且保 留了模型中的细小部件信息。 (2)针对弱纹理立面修复过程中出现的三角面自相交现象，提出抑制拓扑错误预处理 的方法，设计了基于单应矩阵 H 的映射检测算法，进行立面凹陷与凸起检测。剔除凹陷 与凸起的异常三角面，在立面平整度修复完成后通过网修补算法对三维 Mesh 模型的拓 扑结构缺失进行修复。 (3)针对非弱纹理立面不平整、精度低等问题，提出基于区域生长的网分割与超像素 信息结合的方法，提取非弱纹理立面。获取立面后，使用其上的随机采样点集拟合理想 目标平面，并采用网优化的思想，利用理想平面对原始 Mesh 中的非弱纹理立面顶点坐 标进行优化，从而实现非弱纹理立面修复。修复后立面平整度改善效果显著。 为验证本文方法的有效性与可行性，利用北川市和合庆市的无人机倾斜摄影数据进 行实验。结果表明本文方法在对复杂多噪的城市建筑三维模型进行提取与修复时，保留 了立面中细小部件信息，立面平整度得到明显提升，效率高，效果好，弱纹理立面漏检 率均低于 5%，且与商业软件 CC 相比，非弱纹理立面修复后立面平整度提升最高可达 15.5%。 

Real scene 3D modeling has the characteristics of being intuitive, measurable, calculable, and rich in information. It is an important application technology in the process of smart earth construction. Oblique photography 3D reconstruction has the advantages of covering a wide range of scenes, high efficiency of model generation, and low cost. It is one of the main means of producing 3D models of real scenes. However, due to the limited information obtained from the oblique image, the facades lack of features when multi-view image matching is performed, and fewer point clouds are generated. The above phenomenon causes that the facade structure of the model is incomplete, some facades are concave or convex. In order to improve the realism of the 3D model and meet the needs of various spatial analysis, it is necessary to improve the visualization of the 3D model through repair methods. The existing repair methods can divide into two types: semi-automatic and manual repair. In the semi-automatic method, the existence of a large amount of noise makes the edge features of the house distorted, so it is difficult to completely extract and repair building facades. The manual interaction repair method has high cost and low efficiency, which is not suitable for large scene models. Based on above background, this paper proposes a method for intelligently extracting weak texture facades and non-weak texture facades for the problem of building facades with weak texture and obvious noise, and repair 3D building facades to improve the visualization of 3D models. The main research contents of this thesis are as follows: (1) In view of the concave and unevenness of weak texture facades in 3D models, an intelligent method is used to extract the weak texture areas and repair the facades. The method extracts weak texture pixel information in the image based on the multi-slice information of the oblique image combined with the superpixel segmentation method, and uses the combination of back projection and occlusion judgment to obtain the many-to-one relationship between the pixels and the triangular faces in the 3D mesh model. So as to establishing the connection between the 2D weak texture pixels and the 3D weak texture façade, and then the weak textured facades in the 3D model are repaired by plane projection. After repairing with the method in this paper, the flatness of the facades are obviously improved, and the details of small parts in the model are preserved. (2) In order to avoid the self-intersection of 3D meshes in the repair process, the thesis proposes a preprocessing method to suppress topological error, and a mapping detection algorithm based on the homography matrix H is designed to detect the concave and convex of the facades. Remove the concave and convex triangular which is abnormal. And repair the lack of topological structure of the 3D mesh model through the mesh hole repair algorithm after the facades is repaired. (3) In order to solve the problem of unevenness and low precision of non-weak textured facades, the thesis proposes the method of extracting non-weak textured facades based on the combination of region growing mesh segmentation and superpixel information. After obtaining the façade, use the random sampling point set in it to fit the ideal target plane, and adopt the idea of mesh optimization, use the ideal plane to optimize vertex coordinates of the non-weak texture façade in the original Mesh, so as to achieve facades repair. After the repair, the flatness of the façade has been improved significantly. In order to verify the effectiveness and feasibility of the method in this paper, experiments were carried out using the oblique photography data of UAVs in Beichuan City and Heqing City. The experiments show that the repair method in this paper has the characteristics of high efficiency and good repair effect for complex and noisy 3D models of urban buildings, and retains the details information of the small components after repair, the detection of the weak texture façade lower than 5%, and compared with the commercial software CC, the flatness of the facade is improved by 15.5% after repair. 

[1]李德仁. 基于数字孪生的智慧城市[J]. 互联网天地, 2021(07): 12.

[2]朱庆, 张利国, 丁雨淋, 等. 从实景三维建模到数字孪生建模[J/OL]. 测绘学报: 1-13.

[3]林万荣. 倾斜摄影实景三维建模技术及其应用现状分析[J]. 测绘与空间地理信息,

2022, 45(01): 209-211.

[4]郑月，陈吉，孙韬，等. 基于无人机倾斜影像的模型重建单体化及精度研究[J]. 中阿科技论坛(中英文), 2020(10): 83-85.

[5]张祥, 杨剑, 吴浩, 等. 利用多尺度张量投票的建筑立面分割方法[J]. 武汉大学学报(信息科学版), 2021,46(03): 348-352.

[6]李兴久. 浅谈无人机倾斜摄影技术在城市实景三维建模中的应用[J]. 测绘标准化,

2021, 37(01): 75-78.

[7]尚琪森, 朱庆. 特征线约束的倾斜摄影测量三维模型棱线修复[D]. 西南交通大学,

2019.

[8]武凯, 周佳新, 程章林. 基于斐波那契采样的三维线云模型平面提取算法[J]. 集成技术, 2021, 10(06): 74-85.

[9]宋锦阳, 钱建国, 汤圣君, 等. 三维点云室内平面要素提取与优化[J]. 测绘通报,

2021(09): 15-20+27.

[10] 王丽英, 王鑫宁. 多值体素连通区域构建下的机载 LIDAR 数据三维平面提取[J]. 地球信息科学学报, 2021, 23(09): 1598-1607.

[11] 蒯通, 朱庆;胡翰. 轮廓线约束的倾斜摄影测量三维模型立面修复方法[D]. 西南交通大学, 2020.

[12] ZHI Q, TIAN, NAN N, et al. Video object segmentation with shape cue based on spatiotemporal superpixel neighbourhood.[J]. IET Computer Vision, 2014, 8(01):16-25.

[13] YANG F, LU H, YANG M H. Robust Superpixel Tracking[J]. IEEE Transactions on Image Processing A Publication of the IEEE Signal Processing Society, 2014, 23(04): 1639-1651.

[14] MA B, HU H, SHEN J, et al. Generalized Pooling for Robust Object Tracking[J]. IEEE Transactions on Image Processing, 2016, 25(9): 4199-4208.

[15] SHU W, LU H, FAN Y, et al. Superpixel tracking[C]//2011 International Conference on Computer Vision. 2011.

[16] OLSON, CLARK, F., et al. Automatic target recognition by matching oriented edge pixels.[J]. IEEE Transactions on Image Processing, 1997, 6(01): 103-113.

西安科技大学硕士学位论文

[17] COUR T, SHI J. recognizing objects by piecing together the segmentation puzzle[C]//IEEE Conference on Computer Vision and Pattern Recognition. 2007.

[18] LU H, FENG X, LI X, et al. Superpixel level object recognition under local learning framework[J]. Neurocomputing, 2013, 120(23): 203-213.

[19] MORI G, REN X, EFROS A, et al. Recovering human body configurations: combining segmentation and recognition[C]//IEEE Computer Society Conference on Computer Vision and Pattern Recognition. 2004.

[20] FULKERSON B, VE_DA_LDI A, SOATTO S. Class segmentation and object

localization with superpixel neighborhoods[C]//IEEE International Conference on Computer Vision. 2009.

[21] ZHANG, HONGYAN, LI, et al. Efficient Superpixel-Level Multitask Joint Sparse Representation for Hyperspectral Image Classification[J]. IEEE Transactions on Geoscience and Remote Sensing, 2015. 53(10): 5338-5351.

[22] WANG Z, FENG J, YAN S, et al. Image Classification via Object-Aware Holistic Superpixel Selection[J]. IEEE Transactions on Image Processing, 2013, 22(11): 4341-4352.

[23] ZHOU X, CHEN S, CHAN S. Online learning for classification and object tracking with superpixel[C]//IEEE International Conference on Robotics & Biomimetics. 2016.

[24] 马克勤, 杨延娇;李忠. 基于超像素分割的无人机图像建筑物识别[D]. 西北师范大学, 2020.

[25] 宋熙煜. 基于超像素的图像分割技术研究[D]. 解放军信息工程大学, 2015.

[26] 宋熙煜, 周利莉, 李中国, 等. 图像分割中的超像素方法研究综述[J]. 中国图象图形学报, 2015, 20(5): 599-608.

[27] Jianbo Shi, J. Malik. Normalized cuts and image segmentation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2000, 22(08): 888-905.

[28] LIU M Y, TUZEL O, RAMALINGAM S, et al. Entropy rate superpixel

segmentation[C/OL]//CVPR 2011.

[29] ACHANTA R, SHAJI A, SMITH K, et al. SLIC Superpixels Compared to State-of-theArt Superpixel Methods[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2012, 34(11): 2274-2282.

[30] LI Z, CHEN J. Superpixel segmentation using Linear Spectral Clustering[C]//2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). 2015.

[31] CHEN J, LI Z, HUANG B. Linear Spectral Clustering Superpixel[J]. IEEE Transactions on Image Processing, 2017, 26(07): 3317-3330

[32] ACHANTA R, SUSSTRUNK S. Superpixels and Polygons Using Simple Non-iterative Clustering[C]//IEEE Conference on Computer Vision and Pattern Recognition. 2017.

[33] SHEN J, HAO X, LIANG Z, et al. Real-Time Superpixel Segmentation by DBSCAN Clustering Algorithm[J]. IEEE Transactions on Image Processing, 2016, 25(12): 5933-5942.

[34] 林筱韵, 翟羽佳, 吴晓群. 基于"几何-拓扑"迭代优化的三维网格模型修复算法[J]. 计算机辅助设计与图形学学报, 2022, 34(2): 305-314.

[35] 张蜜儿. 顾及刚性形变的街景影像立面纹理纠正方法[D]. 西南交通大学, 2020.

[36] 雷玉珍. 三维点云数据处理中的若干关键技术研究[D]. 华中科技大学, 2013.

[37] 陈寅. 基于泊松方程的浮雕处理技术研究[D]. 国防科学技术大学, 2010.

[38] 张雨农, 李春忠. 密度峰值聚类的优化及其应用研究[D]. 安徽财经大学, 2020.

[39] RODRIGUEZ A, LAIO A. Clustering by fast search and find of density peaks[J]. Science, 2014, 344(6191): 1492-1496.

[40] 张志龙, 李爱华, 李楚为. 基于密度峰值搜索聚类的超像素分割算法[J]. 计算机学报, 2020, 43(01): 1-15.

[41] JUNG Y H, SO B S, LEE K, et al. Swept volume with self-intersection for five-axis ballend milling[J]. Proceedings of the Institution of Mechanical Engineers Part B Journal of

Engineering Manufacture, 2003, 217(08): 1173-1178.

[42] KIM M S, ELBER G, PEKERMAN D. Self-intersection detection and elimination in freeform curves and surfaces[J]. Computer-Aided Design, 2008, 40(02): 150-159.

[43] 张雪波, 方勇纯, 马博军, 等. 基于单应矩阵的摄像机标定方法及应用[J]. 控制工程, 2010, 17(2): 248-251+255.

[44] 童子良, 余学祥, 汪涛, 等. 一种新的鲁棒三维点云平面拟合方法[J]. 合肥工业大学学报(自然科学版), 2020, 43(12): 1684-1688.

[45] FISCHLER M A, BOLLES R C. Random Sample Consensus: A Paradigm for Model Fitting with Applications To Image Analysis and Automated Cartography[J].

Communications of the ACM, 1981, 24(6): 381-395.

[46] 薄怀志. 三种点云数据平面拟合方法的精度比较与分析[J]. 测绘与空间地理信息, 2018, 41(05): 206-208.

[47] 袁天然. 三角网格模型光顺,简化和缝补技术的研究及应用[D]. 南京航空航天大学, 2008.

[48] 李彩玲. 三维模型重建,修复及优化算法研究与实现[D]. 首都师范大学, 2013.

[49] 薛婧雅, 李礼, 龚烨, 等. 一种基于超体素与区域生长的机载点云屋顶平面分割方法

[J]. 测绘地理信息, 2021, 46(S01): 232-236. 西安科技大学硕士学位论文 62

[50] 武新丹, 杨秋翔;孙旭波. 基于区域生长技术的三维模型分割研究与应用[D]. 中北大学, 2020.

[51] VIEIRA M, SHIMADA K. Surface mesh segmentation and smooth surface extraction through region growing[J]. Computer Aided Geometric Design,2005, 22(08): 771-792.

[52] HOPPE H, DEROSE T, DUCHAMP T, et al. Mesh Optimization[C]//Proceedings of the 20st Annual Conference on Computer Graphics and Interactive Techniques, SIGGRAPH 1993.

[53] HOPPE H. Progressive Meshes[J]. Proceedings of the 23rd annual conference on Computer graphics and interactive techniques. 1996: 99-108.