【作者】 陈博；

【导师】 佘江峰；

【作者基本信息】 南京大学 ， 地图学与地理信息系统， 2021， 硕士

【摘要】 三维建筑物是虚拟地理环境最重要的组成元素之一,目前已经广泛应用于三维导航、智慧城市、水利水电等多个领域。随着三维数据采集、三维建模技术的快速发展,三维建筑物模型的精细程度越来越高,数据量也随之急剧增长。虽然近年来计算机性能已得到了飞速提升,但在目前的硬件条件下,大规模三维城市场景的实时流畅渲染仍然面临着比较大的挑战。针对此问题,目前广泛采用的解决方案是细节层次技术(Level of Detail,LOD),其核心是在保持建筑物外观特征的前提下,对三维模型进行不同程度的简化,得到一组不同细节层次的简化模型。在实时浏览时,根据视角、用户到视点的距离等多种因素,动态选择不同精细程度的模型进行加载,以减轻计算机渲染的压力。与地形等流形网格模型相比,三维建筑物模型具有复杂的几何结构,严格的几何约束和拓扑约束,且存在大量非流形网格。将计算机图形学领域的三维模型简化算法直接应用于三维建筑物模型,可能会破坏建筑物固有的几何约束和拓扑约束,造成组件分离、建筑物坍塌等问题,进而严重影响简化模型的外观。为此,国内外学者提出了一系列专门针对三维建筑物的简化算法,这些算法有一部分是对计算机图形学领域网格简化算法的改进,仍存在传统方法的问题;一部分从结构维度对建筑物进行简化,只适用于结构较为简单的建筑物,且结构简化后的模型仍存在网格简化的空间。此外,除了保持模型的外观特征外,简化前后的模型应用于空间分析,得到的分析结论的一致性也非常重要:若简化后的模型无法保持与原模型基本一致的空间分析结论,则可能得到精度不高甚至错误的空间分析结论,这将严重限制简化模型在空间分析中的应用。本研究在国家自然科学基金面上项目“三维建筑物连续LOD模型的生成及其可视化研究”(41871293)的支持下,以保持模型外观和提高模型简化率为目标,提出了一种兼顾模型网格和建筑物结构的三维建筑物简化方法,研究的主要内容和主要结论如下:1)研究三维建筑物的结构提取问题。使用了一种广度优先搜索算法,对三维建筑物模型中的组件进行了提取。为了保证同类组件简化前后的一致性,对同类别的组件根据几何特征进行聚类组合。根据组件间的拓扑关系建立组件关系拓扑图,并根据组件的几何特征和组件间的连接关系将组件或其组合分为主体结构和次级结构,将次级结构再细分为辅助结构、连接结构和组合结构三种结构,最后将这些结构根据组件间的依赖关系用结构树进行组织。2)研究综合考虑模型网格和建筑物结构的三维建筑物简化方法。使用半边折叠算法对三维建筑物模型进行网格简化。根据半边折叠过程中边折叠引起的体积变化越来越大的特点,设置移除阈值,在结构树的约束下,按规则逐步移除结构树中的结构,以实现三维建筑物模型网格和结构的同时简化,避免因为网格过度简化引起的组件脱离、模型坍塌等问题。3)研究从多个角度对简化后的模型进行评价。从外观特征保持和空间分析结论一致性两个方面对简化结果进行评价。外观特征保持方面,使用基于图像的简化质量评价方法进行评价。在三维建筑物模型周围的多个角度设置相机,获取多个视角下模型的二维图像,计算简化后模型与原模型的图像视觉差异,量化比较简化前后模型的外观差异。空间分析结论一致性方面,以阴影分析为例进行空间分析结论一致性的评价。获取简化前后同一地理位置不同时间三维建筑物模型投影在地面上的阴影,比较简化前后模型阴影面积的变化,验证简化前后模型空间分析结论的一致性。实验结果表明,与传统的仅考虑网格简化的算法相比,本研究提出的方法具有更好的简化效果,能有效避免过度网格简化引起的模型破坏问题。同时,在保持简化质量的同时,本研究提出的方法可以达到较高的简化率。此外,高度简化的三维建筑物模型能与原模型在某些空间分析中取得基本一致的分析结论。更多还原

【Abstract】 3D buildings are one of the most important elements of the virtual geographic environment,and they have been widely used in 3D navigation,smart cities,water conservancy and hydropower and other fields.With the maturity of 3D data collection,modeling and visualization technology,the data volume of 3D building models has increased dramatically.Although computer performance has been rapidly improved in recent years,under current hardware conditions,the real-time smooth rendering of large-scale 3D urban scenes still faces a large challenge.To solve this problem,Level of Detail(LOD)technology is widely used.The key of LOD is to simplify the 3D model to different levels while maintaining the appearance of the building and obtain a set of models of different levels of simplification.During the real-time interaction process,according to factors such as the angle of view,the distance from the user to the viewpoint,etc.,a specific model level is selected and loaded according to its distance from the viewpoint.This approach can greatly reduce the burden of transmission and rendering.Compared with manifold mesh models such as terrain,3D buildings usually have complex structures,strict geometric and topological constraints,and many nonmanifold meshes.Applying the 3D model simplification algorithm in the field of computer graphics directly to the 3D building model may cause problems such as component separation and building collapse due to the destruction of the geometric and topological constraints of the building,which seriously affects the appearance of the simplified model.To solve above questions,scholars have proposed a series of simplified algorithms for 3D buildings.Some of these algorithms are aimed at the improvement of traditional mesh simplification algorithms in the field of computer graphics,but there are still problems with traditional methods;The simplification of structures is only suitable for buildings with relatively simple structures,and there is still room for mesh simplification.In addition to maintaining the appearance characteristics of the model,the consistency of the spatial analysis conclusions before and after the simplification is also very important: if the simplified model cannot guarantee the consistency of the spatial analysis conclusions,the simplified model cannot be applied to the spatial analysis to improve the efficiency of spatial analysis.This research is supported by the National Natural Science Foundation of China’s“Study on the Construction and Visualization of Continuous LOD Models of 3D Buildings”(41871293).Aiming to maintain the appearance of the simplified models and improving the simplification rate,a 3D building simplification method that considers both the model mesh and the building structure is proposed.The main contents and conclusions of the study are as follows:1)Study the structure extraction of 3D buildings.We use a breadth-first search algorithm to extract the components in the 3D building model.In order to ensure the consistency of the simplified components of the same category,the separate components of the same category are clustered according to their geometric features.A component topology graph is established according to the topological relationship between the components,and the components and their combinations are divided into the main structure and the secondary structure according to the appearance characteristics of the components and the connection relationship between the components,and the secondary structure is subdivided into auxiliary structures,connection structure and combination structure.After that,these structures are organized by the multi-way data structure named structure tree.2)Research on a 3D building simplification method considering both the model mesh and the building structure.The half-edge collapse algorithm is used to simplify the mesh of the 3D building model.According to the characteristics of the increasing volume change caused by edge collapse during the half-edge collapse process,the removal threshold is set.Under the constraints of the structure tree,the structure in the structure tree is gradually removed according to removal rules.Through this,we realize the simultaneous simplification of the mesh and structure of the 3D building model,which effectively avoid problems such as component separation and model collapse caused by excessive mesh simplification.3)Evaluate the simplified model from the aspects of maintaining appearance characteristics and consistency of spatial analysis conclusions.We set up several cameras around 3D building model from multiple angles,obtain images of the model from multiple perspectives,use image-based simplified quality evaluation methods to calculate the visual difference between the simplified model and the original model,and quantify the comparison of the appearance of the model before and after the simplified model.In the consistency of spatial analysis conclusions,shadow analysis is taken as an example to evaluate the consistency of spatial analysis conclusions.By obtaining the shadow of the 3D building model projected on the ground at the same location and at different times before and after the simplification,the changes of the shadow area before and after the simplification are compared,and the consistency of the model space analysis conclusions before and after the simplification is verified.Experimental results show that,compared with traditional methods,the method proposed in this study has better simplification results than traditional methods that only consider mesh simplification,and can effectively avoid model destruction caused by excessive mesh simplification.At the same time,while maintaining the quality of simplification,the method proposed in this study can achieve a higher simplification rate.In addition,the highly simplified 3D building model can reach a basically consistent conclusion with the original model in some spatial analysis.更多还原