【作者】 吴宁；

【导师】 李王鸣； 陈秋晓；

【作者基本信息】 浙江大学 ， 建筑设计及其理论， 2013， 博士

【摘要】 随着数字城市的快速发展和城市三维空间模型应用领域的不断扩展,城市建筑群三维模型的需求日益增长。然而传统建筑物三维重建方法在面对大空间尺度、大数据量、更新节奏快的城市建筑群时,在效率、精度、成本、尺度、技术门槛等方面均不同程度存在缺陷。为此,论文尝试引入目标识别和参数化技术,探求一种适用于城市建筑群的低成本、低门槛、高效率的“大众化”三维重建解决方案,以满足数字城市及相关领域的迫切需求。论文的主要内容包括：(1)对建筑物三维重建、目标识别和参数化建模这三大领域的相关研究、技术方法作了总结回顾,指出了其存在的问题；(2)对三大技术体系作了深度解构,指出了“二元并行框架”的成因,给出了交叉的可行途径,并通过体系重构,构建了城市建筑群三维重建的“三元交叉框架”；(3)针对遥感影像分割,提出了面向对象的多尺度区域合并分割方法和基于量化合并代价的快速区域合并分割方法；(4)针对矢量图形优化,提出了基于删除代价的矢量图形单层次优化方法、面向遥感影像矢量化图形的多层次优化方法和面向建筑群的矩形拟合优化方法；(5)针对三维信息提取,提出了基于扩展统计模型的建筑群高度提取方法和三种城市建筑群层数估算模型,针对侧向航拍影像提出了一种建筑群坐标修正方法；(6)提出了“参-建分离”的系统架构。针对该架构中的参数管理模块,设计了参数的关联、组织、管理和属性块的恢复、管理等方法。针对服务网站模块,设计了风格库管理和项目库管理子模块。针对自动建模模块,提出了DXF-SHP文件格式自动转换,CGA文法规则和规则库框架设计,规则库的调用和参数值传递,以及自动化建模脚本的设计等系列方法；(7)集成关键技术方法,开发了城市建筑群三维重建软件原型系统(3DRS)及其子系统CBRS、CityUp,并以杭州市西湖区为案例开展了实证研究,从精度、效率、成本、技术门槛、时效性等方面验证了整套解决方案的可行性。研究表明：(1)建筑物目标识别与参数化建模技术,恰可解决建筑物三维重建面临的两大难点；“三元交叉框架”使三大技术体系成为一个紧密连接、流程清晰、分工明确、目标一致的统一整体,为相关研究提供了理论支撑和方法指导；(2)面向对象的多尺度区域合并分割方法使综合考虑多种地物特征和多尺度分割成为可能,最大限度地缩短了队列长度,提高了分割精度和速度；基于量化合并代价的快速区域合并分割方法不仅能够保证分割精度,而且合并速度优势随初始分割区域数量的增加而越加显著；(3)相较于经典DP方法,基于删除代价的矢量图形单层次优化方法具有更高的精度和更低的时间复杂度,而且单位节点处理能力和等压缩率下的处理速度均具有显著优势；相较于传统单层次优化方法,面向遥感影像矢量化图形的多层次优化方法对影像不同的分割尺度和空间分辨率具有更强的适应性,能更好地还原地物的多层次特性；面向建筑群的矩形拟合优化方法可以有效减少最小面积外接矩形的计算时间,确保矩形对边平行、邻边垂直的关系,所得优化结果的形态和面积精度均较为理想；(4)基于扩展统计模型的建筑群高度提取方法、建筑群层数估算模型和建筑群坐标修正方法分别达到了较高的精度水平,可满足相关应用需求；(5)“参-建分离”的系统架构以及针对三大模块提出的一系列创新方法大幅降低了参数化建模平台的技术门槛和边际成本,提高了建模效率,为参数化技术的快速、广泛普及提供了新的发展思路；(6)本研究所提方法及系统在精度、效率、成本、门槛、时效性等方面均满足实验预定目标,体现了广泛的优势和大众化特性,整套解决方案具有可行性；(7)研究成果将推动数字城市三维空间数据基础设施的建设,使之在城市规划与管理、建筑景观设计、国防军事、应急救灾、环境保护、虚拟旅游、交通导航等诸多领域得到更广泛的应用。更多还原

【Abstract】 With the rapid development of digital city and the constant expansion of city3D space model application fields, demand in the3D model of city buildings grows day by day. Nevertheless, facing city buildings with large spatial scale, huge data volume and quick data update demand, traditional buildings3D reconstruction method has some disadvantage in terms of efficiency, precision, cost, scale, technical threshold, etc. Hence, this essay attempted to introduce target recognition and parameterization technology, and presented a popular3D reconstruction solution with low cost, low threshold and high efficiency that adapt to city buildings, thus met the urgent demand for digital cities and related fields.The main content of this essay is as follows. First, it summarized and reviewed three main fields including building3D reconstruction, target recognition and parametric modeling, and pointed out the existing problems of the above fields. Second, three major technology systems were deconstructed in depth and causes of "dual parallel framework" were pointed out as well. It also offered a practical way of the cross."Ternary cross framework" of3D reconstruction in city buildings was constructed through system reconfiguration. Third, aiming at remote sensing image segmentation, it also presents an object oriented multi-scale region merging method and a fast region merging method based on the classified merging cost. Fourth, aiming at vector graphics optimization, it presented a single-level optimization method for vector graphics based on deleting cost, a multi-level optimization method for remote sensing image vector graphics as well as a rectangle fitting optimization method for the building groups. Fifth, aiming at3D information extraction, it presents building height extraction method based on extended statistical model and three models to retrieve building storey information. A geometry correction method for building footprints aiming at lateral satellite images was also put forward. Sixth, a "parameter-modeling separation" system framework was presented. Aiming at parameter management module of this framework, the method of parameter association, organization and management as well as recovery and management of attribute block of CAD were designed. Composed of style library management and project library management sub-module, web site service module was designed. Aiming at automatic modeling module, DXF-SHP conversion, the design of the CGA syntax rules, rule library frame, rule library calling and parameter transmission as well as automatic modeling script were put forward. Seventh, through integrating key technology and methods, it also developed city buildings3D reconstruction software prototype system (3DRS) as well as its subsystem CBRS and CityUp. Meanwhile, empirical research was developed based on the example of the West Lake district in Hangzhou city. Also, the feasibility of the whole solutions was evaluated from the following aspects like precision, efficiency, cost, technical threshold, data presentalism, etc.Research showed that, first, building target recognition and parametric modeling technology could exactly solve two difficult points of building3D reconstruction."Ternary cross framework" made three technology systems form a closely connected, procedure clear, division identified and target consistent unit, which offers theory support and method guidance for relevant research. Second, object oriented multi-scale region merging method made multi-scale segmentation in consideration of terrain features possible. By reducing the queue length to a large extent, segmentation precision and implementation speed was improved. Not only does the fast region merging method based on classified merging cost guarantee segmentation precision but also merging speed advantage becomes notable with the increase of the initial segmented region number. Third, compared with classical DP method, the single-level optimization method for vector graphics based on deleting cost was of higher precision and lower time complexity. Also, it outperformed the former in terms of processing capability per unit node and processing speed at the same compression ratio. Compared with the traditional single-level optimization method, multi-level optimization method for remote sensing image vector graphics had stronger adaptability to image segmentation scale and spatial resolution, which can better restore multilevel character of ground feature. Rectangle fitting optimization method for building groups could effectively reduce computation time of the minimum area bounding rectangle, ensured rectangles be parallel on opposite side and be perpendicular on adjacent side. The shape and area precision of the optimization result were both relatively ideal. Fourth, three methods including the building height extraction method based on extended statistical model, the building storey information retrieval method and the building footprint geometry correction method, respectively reached higher precision level and could meet corresponding application demand. Fifth, the "parameter-modeling separation" system framework and the series of innovation methods aiming at the three modules largely reduced technical threshold and marginal cost of parametric modeling platform and increased modeling efficiency, thus offering a brand-new strategy for the quick and widespread popularity of parameter technology. Sixth, methods and system proposed in this essay met the experiment’s preseted target in terms of precision, efficiency, cost, threshold, data presentalism, etc. It showed comprehensive advantages with public orientation. The whole solutions are feasible. Seventh, it is very likely the research results will push forward the construction of3D digital city spatial data infrastructure, making it more widely used in urban planning and management, architecture and landscape design, national defense and military, emergency relief, environmental protection, virtual tourism, transportation navigation, etc.更多还原