>> 王庆国,2007,三维城市建筑群模型的自动综合方法
摘 要
近年来,随着城市信息化建设的快速发展,对三维城市模型(3D City Model,3DCM)的研究成为一个普遍关注的前沿性课题,3DCM也在越来越多的领域得到越来越广泛的应用。
但是,3DCM数据的几何复杂度高、纹理数据量庞大,现有的计算机硬件难以实现对模型的实时交互显示。因而,通常需要在不影响浏览真实感的前提下,降低城市数据的几何复杂度和纹理数据的大小,提高相关数据的绘制速度。另一方面,对于3DCM的不同应用领域而言,其对模型的精细程度和信息的详细程度、对模型表达的真实感和空间范围等方面的要求也都存在着显著差异。有些应用需要较为精细的模型和较为详细的细节信息,而另一些应用却只需要较为粗略的模型和较为概括的信息;有些应用需要具有相片质感的可视化感受,而另一些应用只需要能够反映出城市的整体结构特征即可。因此,在三维城市模型的发展过程中,多细节层次是其最重要的特征。而在三维城市模型中,建筑物模型是其最重要的组成部分,所以,对三维城市模型多细节层次表达问题的研究,一般集中于对建筑物的描述。
实现模型多细节层次表达的一种最理想方法就是以某一细节层次程度最高的模型为基础,通过模型综合的方法,根据需要派生出相应的低细节层次模型。对三维城市建筑物模型的综合,包括对单个建筑物模型进行简化和对建筑群模型进行综合两个方面,但是现有的研究主要都集中在对单个建筑物三维模型的自动简化方面,对由邻近建筑物构成的建筑群的三维模型的综合则较少涉及。本研究就是针对城市局部范围内的三维建筑群模型的综合问题展开研究,主要研究内容包括:
(1) 全面分析了二维地图综合和三维模型简化的理论与方法,归纳了综合问题研究的发展趋势:从二维地图综合向三维模型综合发展;从手工综合向借助计算机的自动综合发展;从仅仅服务于可视化目的向服务于多种目的发展;从对简单要素的综合向对复杂的多维形体和场景的综合发展。在此基础上,确立了本文的研究主题:城市局部范围内的邻近三维建筑物几何模型的自动综合。
(2) 提出了一种三维城市建筑群模型的层次分割方法。研究内容包括:城市建筑群特征的分析;相关城市理论(城市空间结构理论、城市形态理论、城市意象理论)分析及其对城市建筑群模型综合的指导意义的挖掘;三维城市建筑群模型的分割标准和分割概念方法的建立。最后以道路分割为例,建立了三维表达的情况下城市建筑群模型基于道路的层次分割方法。不同于二维地图综合中对道路的选取只需要考虑道路自身的重要性,在三维表达的情况下,对道路的选取不仅需要考虑道路自身的重要性,还需要考虑道路在视场中的位置,即与视点的相对位置关系。在视觉敏感的区域中,一些可明确分辨的重要性等级不高的道路可能也要选取,使分割的程度更精细。而在视觉不敏感的区域,如果不能明确辨识,即使是重要性等级较高的道路,可能也不选取,分割的程度比较粗糙。
(3) 提出了基于邻近性和相似性的空间聚类方法。在研究三维建筑群模型分布结构特征的分析和识别基础上。针对传统聚类算法将对象抽象成无大小和无方向的点,聚类中仅仅考虑对象之间的欧氏距离的局限,通过实验证明三维建筑群分布结构的聚类分析中至少要考虑邻近性和相似性两个方面的因素,突破了现有的空间聚类方法仅仅考虑邻近性的局限。
(4) 针对三维建筑群几何模型的综合,建立了从几何特征约束,到关系特征约束,再到分布结构特征约束三个层次的多重约束模型。在三个层次中,几何特征约束处于最底层,直接受到细节层次变化的影响;分布结构特征的约束处于最高层,是在综合过程中必须始终保持的。高层约束对低层约束具有指导和控制作用。
(5) 基于尺度空间的思想,提出了一种建筑物聚合操作算法,并综合应用前面几章的研究成果,以聚合操作为基础,进行了直线分布建筑群的综合实验,实现了对三维建筑群模型的综合,证明了算法的有效性。
关键词:
尺度;多细节层次;三维城市模型;三维城市建筑群模型;自动综合;特征
Abstract
With the rapid development of city information construction, research on 3D City Model (3DCM) has become a hot and widely attractive issue. Moreover, 3DCM is widely applying in many fields.
However, the geometric structure of 3DCM is complicated and the volume of texture data is huge, which is beyond the real time rendering abilities of current work stations. A common solution is to decrease the data volumes of geometric data and texture data to improve rendering speed. On the other hand, various applications have different requirements on the accuracy of geometric data and visual effects. For example, a lot of applications may emphasis on the details of geometric data. Some applications may need a holistic structure. Hence, the levels of detail (LoD) of 3DCM is of vital importance for a 3DCM, which has attracted wide research in recent years.
An ideal solution of generating Lod models is to derive a set of low resolution Lods from an original model based on generalizations. The generalization of 3DCM includes the simplification of single building model and the generalization of 3D urban building group models. However, existing research mainly focuses on the former; the latter one receives less attention. This thesis aims at the generalization of 3D building group models in local urban area, the main contents include:
1) Firstly, the theories and approaches in 2D map generalization and 3D model simplification are reviewed. Secondly, the trend of map generalization is summarized, namely, from 2D map generalization to 3D model generalization, from manual generalization to automated generalization, from visualization oriented to multi-applications oriented, from generalization of simple elements to generalization of complicate multi-dimension objects and scenes. Then, the main research topic on the thesis, automated generalization of adjacent 3D building geometric models in local urban area, is built.
2) The thesis proposes a hierarchical partition of 3D urban building group model in Chapter 2. Firstly, the characteristics of urban building groups are analyzed, correlative theories on city are analyzed and mined for the generalization of urban building group models. Then, the criterions and approaches of partition are established. Finally, a partition approach based on road is implemented for the partition of 3D urban building group models, which is quite different from the partition approach for 2D.
3) Chapter 3 presents a clustering approach based on adjacency and similarity. In the traditional clustering approaches, objects are treated as points without sizes and orientation, and only the adjacency between objects is considered. However, both the adjacency and similarity of objects should at least be taken into account for 3D clustering.
4) Chapter 4 presents generalization constraints, which aim at the generalization of 3D building group models. Three layers of constraints, namely, geometric constraints, relation constraints, and distribution structure constraints, are concerned.
5) Chapter 5 shows aggregation algorithm experiments. Based on the idea of scale-space, an aggregation algorithm is established and proved to be valid for the generalization of 3D building group models.
Key words: Scale; Levels of Detail; 3D City Model; 3D Urban Building Group Model; Automatic Generalization; Characteristic