EN中文

地球信息科学学报 >

2016 , Vol. 18 >Issue 3: 406 - 415

DOI: https://doi.org/10.3724/SP.J.1047.2016.00406

地球信息科学理论与方法

基于房屋轮廓与纹理的三维建筑模型分层次聚类研究

  • 潘文斌 , 1, 2, 3, * ,
  • 刘坡 1, 4 ,
  • 周洁萍 2 ,
  • 龚建华 2
展开
  • 1. 城市空间信息工程北京市重点实验室,北京 100038
  • 2. 中国科学院遥感与数字地球研究所 遥感科学国家重点实验室,北京 100101
  • 3. 中国科学院大学,北京 100049
  • 4. 中国测绘科学研究院,北京100830

作者简介:潘文斌(1990-),男,硕士生,研究方向为虚拟地理环境.E-mail:

收稿日期: 2015-05-08

  要求修回日期: 2015-06-24

  网络出版日期: 2016-03-10

基金资助

国家自然科学基金面上项目(41371387)

城市空间信息工程北京市重点实验室开放研究基金项目(2014210)

收起

3D Building Model Hierarchical Generalization Based on Building Contour and Texture

  • PAN Wenbin , 1, 2, 3, * ,
  • LIU Po 1, 4 ,
  • ZHOU Jieping 2 ,
  • GONG Jianhua 2
Expand
  • 1. Beijing Key Laboratory of Urban Spatial Information Engineering, Bejing 100038, China
  • 2. State Key Laboratory of Remote Sensing Science, Institute of Remote Sensing and Digital Earth (CAS), Beijing 100101, China
  • 3. University of Chinese Academy of Sciences, Bejing 100049, China
  • 4. Chinese Academy of Surveying & Mapping, Beijing 100039, China;
*Corresponding author: PAN Wenbin, E-mail:

Received date: 2015-05-08

  Request revised date: 2015-06-24

  Online published: 2016-03-10

Copyright

《地球信息科学学报》编辑部 所有
Fold

摘要

目前三维建筑模型已广泛应用于城市规划,导航和虚拟地理环境等领域.不同细节的模型是LOD( Level of detail )技术的基础,由于三维模型的生产成本高昂,模型自动化简逐渐引起了学者的关注.三维模型化简包括单模型化简和多模型综合2方面,目前单个模型的化简研究比较多,而模型群组综合的研究仍然处于起步阶段.本文主要研究模型群组的聚类综合,提出一种基于房屋轮廓与纹理的分层次聚类算法:首先,基于房屋的底面轮廓构建约束Delaunay三角网,以道路为基准对三角网进行划分,通过可视分析构建初始的邻接图,使建筑群组分类符合城市形态学;其次,将房屋纹理引入三维模型群聚类的过程,使用SOM( Self-organizing Map )智能分类算法对纹理进行分析,然后分割邻接图;最后,以最邻近距离对邻接图构造最小生成树,并进行线性检测,将离散的建筑合并到已聚类的群组中,最终完成模型的合并.本文利用纹理辅助轮廓特征,实现三维建筑模型的聚类,符合人类的视觉习惯,实验结果证明了本文方法的有效性.

关键词： 三维模型; 综合; 房屋轮廓; 纹理; SOM

本文引用格式

潘文斌 , 刘坡 , 周洁萍 , 龚建华 . 基于房屋轮廓与纹理的三维建筑模型分层次聚类研究[J]. 地球信息科学学报, 2016 , 18(3) : 406 -415 . DOI: 10.3724/SP.J.1047.2016.00406

Abstract

As a major factor of smart city, 3D building model has been widely applied in many fields such as city planning, navigation and virtual geographic environments. Level of detail (LOD) technology is often used to visualize these models in complex urban environment. In order to reduce the high production cost, 3D building model generalization is gradually attracting attention. 3D building model generalization includes single model simplification, which has been studied widely and deeply, and multiple models synthesis, which is still at its early stages. In this paper, a self-organizing map (SOM) based classification algorithm is used to analyze the texture, and then the texture is combined with geometry feature to generalize 3D building models. In addition, a cognition-based hierarchical clustering algorithm is proposed. Firstly, a constrained Delaunay triangulation (CDT) is constructed from the 2D footprints of building models, and a connected network is established by visibility analysis based on the CDT, which is divided based on the road network to make it conform to the urban morphology theory. An initial proximity graph is constructed from the CDT. Secondly, the proximity graph is segmented according to the texture feature. Finally, a minimum spanning tree (MST) is created from the proximity graphic based on the minimum distance, and after linear detection and discrete model conflation, the model groups are conflated and visualized. The texture feature is used to synthesize 3D model that is adaptive to people's cognitive habits. The experimental results confirm the effectiveness of this method.

Key words： 3D building models; generalization; building contour; texture; SOM

1 引言

三维建筑模型在城市规划,导航和虚拟地理环境等领域得到广泛应用[1].近年来,硬件发展非常迅速,大规模城市三维模型渲染不再是难点,然而多细节层级(Level of Detail,LOD)技术,仍被广泛应用于三维模型可视化,特别是移动导航和旅游领域[2].当视点较远时,使用简单模型代替复杂模型,能减少数据量并提高渲染速度[3].目前,三维模型化简逐渐引起学者的关注,其主要包括单个模型 化简和多个模型综合[4]2个方面.单个模型的化简研究比较多[3,5-6],而模型群组聚类综合的研究还处于起步状态,故本文主要研究模型群组的聚类综合问题.
三维建筑模型投影到底面可构成二维地图,但在二维制图综合中通常不考虑高度,纹理等重要的三维建筑信息,因此二维地图的综合方法需进行扩展才能应用于三维模型综合.三维模型主要由几何网格和纹理组成(包括几何,纹理和语义特征),而三维模型群组综合方法主要基于建筑的几何特征.对此,Chang等[7]基于相邻建筑之间的欧式距离,提出了一种单链接方法对模型群组进行聚类,同时保持了城市的主要结构特征[8].Yang等[9]考虑建筑之间的方向,对单链接聚类方法进行改进,提升了算法的效果.Kada[2]同样基于相邻建筑之间的距离,提出一种基于形态学操作的混合方法,对模型群组进行聚类.为了保持建筑群的结构特征,Regnauld[10]提出一种基于MST和邻接图的二维地图聚类方法,并通过从Gestalt原理推导出的准则对邻接图进行分割.基于城市形态学和Gestalt原理,Zhang等[11]提出一种基于认知的方法对三维模型进行聚类和综合,使聚类结果更符合人类认知习惯.三维聚类是一个复杂的问题,Guercke等[12]将综合当作一个混合整数规划问题,通过求最优解来实现模型的聚类.这些综合方法主要是基于几何特征(如高度,面积,体积和方向),很少考虑建筑底面轮廓进行模型聚类的研究.
纹理特征与人类的空间视觉认知密切相关,基于图像的综合更能保持城市意象[4].目前一些学者在三维建筑模型的聚类研究中,采用符合人类空间视觉认知的纹理进行探讨.Chang等[13]提出一种层次纹理方法,将每个聚类模型外墙和屋顶的纹理重新投影拉伸到集群的高度,可保持模型的视觉特征.Glander使用生成的非结构化图像块代替原始三维模型,根据相机远近来动态突出显示地标[14-15].Zhang等[16]虽然实现了纹理和几何结构的统一管理,但在三维建筑模型的聚类过程中,并没有充分利用纹理信息.这些研究均集中于提取多尺度的三维建筑纹理特征,旨在对三维建筑模型聚类后的结果进行更符合人类空间视觉认知的可视化表达,但对纹理信息在三维建筑模型的聚类过程本身的作用没有深入探讨.
上述研究方法在三维建筑聚类研究中,往往将三维建筑聚类与聚类后的可视化效果分割开来,一部分集中于研究三维模型的几何特征,进行三维建筑模型聚类;另一部分集中于研究在聚类之后,将纹理附着在三维模型上,使其符合人类的空间视觉认知.本文将纹理特性引入三维模型群组的综合过程,结合建筑轮廓特征,对三维建筑模型进行聚类和化简,提出了一种基于房屋轮廓与纹理的三维建筑模型分层次聚类方法.

2 基于房屋轮廓与纹理的三维建筑 模型聚类方法

三维建筑模型聚类综合主要有数据预处理,模型聚类,模型综合与可视化3部分内容.以房屋轮廓与纹理进行聚类时,需准备道路网,房屋轮廓,房屋纹理等数据(图1).首先,基于建筑模型的二维底面轮廓构造约束Delaunay三角网(Constrained Delaunay Triangulation,CDT),并使用道路网初步分割CDT,使之符合城市形态学特征;然后,通过可视分析构造初步的邻接图,并采用SOM(Self-Organizing Map)聚类方法对纹理进行分类,对初始的邻接图进一步分割;最后,基于邻接图构造最小生成树(Minimum Spanning Tree,MST),依次进行线性检测和离散面合并,以Delaunay三角网对模型合并和可视化,最终完成三维模型聚类.
Fig. 1 The framework of the algorithm

图1 算法的主要框架

2.1 基于房屋轮廓的邻接图构建

构建邻接图主要包括:(1)生成CDT并用路网分割三角网;(2)通过分析三角网中每个三角形,构造初始邻接图G1.
2.1.1 约束Delaunay三角网
Delaunay三角网除了端点,平面图中的边不包含点集中的任何点且没有相交边,广泛应用于图形综合[11-17].传统的Delaunay三角网一般是基于建筑重心构造的,其边很难反映建筑的合并关系.如图2(a)左下角表示,在模型合并的过程中,面a与d或者面a与b可直接合并,而面a与e不可能直接合并,面e与c也类似.
Fig. 2 Delaunay triangulation

图2 Delaunay三角网

本文将建筑的底面轮廓作为约束边嵌入Delaunay三角网得到CDT.若一个建筑轮廓的底面顶点序列为<Vi1,Vi2,, et al.,Vij>,则所有建筑轮廓顶点构成的顶点集合V={V11,V12,, et al.,V1m,V21,V22,, et al.,V2n,, et al.,Vk1,Vk2,, et al.,Vks},所有建筑轮廓边构成的约束边集合C={<V11V12,V12V13,, et al.,V1m-1V1m,V1mV11>,<V21V22,V22V23,, et al.,V2n-1V2n,V2nV21>,, et al.,<Vk1Vk2,Vk2Vk3,, et al.,Vks-1Vks,VksVk1>}.基于顶点集合V构建Delaunay三角网,而后插入约束边集合C调整三角网.由此构成的初始CDT1中的三角形分3部分:(1)三角形完全位于建筑轮廓构成的封闭多边形内部;(2)三角形处于建筑轮廓外部且3个顶点位于不同的建筑轮廓上;(3)有极少的三角形位于建筑轮廓外部,但3个顶点位于同一个建筑轮廓,这是由于该建筑轮廓为凹多边形.遍历CDT1中的三角形,剔除第(1)部分和第(3)部分三角形,保留第(2)部分三角形,生成新的CDT2,如图2(b)所示.连接不同建筑的三角形可表示不同建筑间潜在的邻近关系.
交通和水系轴线是城市的基础骨架,在全局范围使用道路网对城市模型进行划分,可保持城市总体形态,布局和意象,保留城市中的街区,道路网,交叉口和地标等主要特征[8-18].因此,需使用城市道路网对CDT2进行分割:遍历CDT2中的三角形,判断其边与道路的相交关系,如果三角形任意一条边与道路网相交,则删除该三角形.删除相交三角形之后的三角网CDT3如图2(b)所示,图中虚线表示被删除的三角形.如果不存在道路数据,则可通过栅格距离来提取道路网[16].通过分析CDT3可提取建筑之间可能的合并关系.
2.1.2 构建邻接图
CDT3中的三角形按其连接的建筑轮廓数目可以分为2类,1类是三角形与2个轮廓相连;2类是三角形与3个轮廓相连.在三维建筑模型合并的过程中,一般是两两合并,因此,当CDT3中的三角形是第1类时,该三角形连接的2个建筑相邻,存在合并的可能.遍历CDT3三角形,可提取可能的合并关系,并构造出初始邻接图.假设有m个建筑,用Oi代表i建筑,则算法步骤如下:
图3(a)中虚线表示图2(b)中被保留的约束三角网,图中三角形分为1类和2类.在聚类的过程中,应剔除第2类三角形,图3(b)表示剔除之后的三角网.基于CDT3构造的初始邻接图G1如图3(c)所示.其中,链接点代表对应的建筑,连接边表示面之间可能直接合并,后续的分类都基于G1.此外,本算法中假设最邻近距离小于2 m的模型本身是合并在一起的,不参与后续分类,如图3(c)左上角的面56,57和58.
Fig. 3 Proximity graph

图3 邻接图

2.2 基于SOM的房屋纹理聚类

纹理本身是有规律和层次的,包含方向,位置和形状等结构特征,如同批建造或结构功能相似的建筑可能具有相同纹理.纹理主要包括屋顶和侧面纹理,在模型合并的过程中,会丢失部分侧面纹理,本文主要研究屋顶纹理.纹理特征是一个难以量化的变量,纹理分析主要包括统计分析法和结构分析法,统计分析法主要统计图形的属性特性,主要包括纹理区域的灰度直方图,灰度平均值,标准差,最小和最大像元值.纹理结构分析主要研究组成纹理的基元及其排列规则,从结构上探索纹理的结构规律,本文重点研究纹理的统计特征[19],采用SOM智能算法,对纹理进行自动分类[20],以每个房屋纹理的平均值和标准差为算法的输入值.
首先计算每个建筑顶部纹理的灰度平均值,如果一个屋顶纹理包含 n 个像元,每个像元的颜色值为 R i 、 G i 、 B i ,则颜色平均值 ( R ̅ , G ̅ , B ̅ ) 如式(1)所示.
R ̅ = ∑ i = 1 n R i n G ̅ = ∑ i = 1 n G i n B ̅ = ∑ i = 1 n B i n (1)
为了提高分类精度,标准差指标也经常被用于分类,纹理标准差指标 σ 如式(2)所示.
σ = ( R i - R ̅ ) 2 + ( G i - G ̅ ) 2 + ( B i - B ̅ ) 2 n (2)
图4是用计算出的顶部纹理平均值填充建筑底面轮廓的示意图,数字表示建筑编号.图中建筑10和11的几何特征(面积,大小和方向等)很相似,其纹理灰度平均值明显不同,使用纹理特征比几何特征更容易将其区分.
Fig. 4 Texture analysis

图4 纹理分析

i 号建筑的纹理平均值和标准差构成的向量 X i = [ R ̅ i , G ̅ i , B ̅ i , σ i ] ,其中各分量的量纲相同.将 X 1 X 2 … X m 等多个向量作为SOM神经网络的输入值,因此,SOM输入层的神经元个数是4,根据区域内颜色数目设定SOM输出层结构为2×2,即2行2列,训练迭代次数为200次.纹理是一个相对特征,还需要选择合适的分类区域,利用路网对三维模型初步分类之后,在子区域进行纹理分析,可提高聚类的精度.
图5(a)为对全部三维模型屋顶纹理进行分类的结果,图5(b)为通过路网将模型分为不同的区域,再对小区域的纹理进行分类的结果.图中用M-N表示建筑,M表示建筑的编号,N表示分类之后的纹理类别.通过观察可明显看出,建筑46和47 的纹理属于一个类别,而使用全局分类则出现了错误.
Fig. 5 Texture classification

图5 纹理分类

2.3 邻接图分割

以像元平均值和标准差指标,使用SOM方法对纹理聚类后,需对初始邻接图进行处理.如果一个边连接的2个轮廓属于不同的纹理类别(图5(b)),则删除该边.图6表示G1经过纹理分割之后的邻接图G2.如图6所示,通过纹理特征将建筑群49,50,51,52和建筑群21,22,23,63分为不同的类别.通过面积和方向,很容易将51,50和49聚类,而将面52分为另外的一组,但是它们的纹理特征几乎一致,在视觉空间中更倾向于聚为一组.
Fig. 6 The proximity graph G2 after texture classification

图6 纹理分类之后的邻接图G2

与纹理类似,遍历G2中的连接边,计算该边连接的2个建筑的面积差,高度差,方向差,如果超过一定阈值则将该链接断开.如果G2中连接边的权重超过阈值,表明该边连接的2个建筑相隔距离较远,则断开链接,最后形成新的邻接图G3.

2.4 三维建筑模型聚类

纹理分析后需对建筑进行构建最小生成树,线性检测和离散建筑聚类等分析.
2.4.1 最小生成树
最小生成树是多边形群的连接图,具有所有连接点相通,无闭合环且树的连接边距离和最小的特点,通过树的逐级剪枝可获得不同层次的聚类结果,已广泛应用于建筑聚类[21].本文基于最邻近距离,对分割后的邻接图构成最小生成树(图7),其链接点为建筑的重心,边的权重为建筑之间的最邻近距离.
Fig. 7 MST (Minimum Spanning Tree)

图7 最小生成树

2.4.2 线性检测
在制图综合中,距离最近的建筑不一定适于合并,还要考虑其他的特征,如相邻建筑之间的面积差,高度差,相似性差,方向差,线性差,间隔差和倾斜度等.在视觉识别中,一般倾向于将同一个方向上的建筑合并为同一类[17],因此还需对最小生成树进行线性检测.
图8(a)表示经过应用纹理和Gestalt准则分割后的邻接图G3,图8(b)表示对应的最小生成树.线性检测的主要步骤如下:(1)从链接点a开始,ab为起始链接,bc为其下一个链接,如果ab和bc的方向差异较大,则断开bc链接;(2)从图8(a)查找与链接点b可能的链接关系,如果be链接与ab的方向相似,则连接be;(3)处理完链接点b之后,处理MST的下一个链接点c,重复步骤(1)和(2),直到遍历整个最小生成树.图8(c)表示线性检测之后的邻接图G4.
Fig. 8 Linear detection

图8 线性检测

2.4.3 离散面链接
模型聚类后,还需进一步处理离散的模型[22],将离散模型合并到已有的群组,主要的步骤:(1)遍历每个离散面,通过CDT构造的初始邻接图G1判断每个面可能的链接关系;(2)通过线性检测将离散面合并到现有的群组.
如图9中的离散面g和h,面h在邻接图中与面e和f链接,需要计算面h与e的接线he与eb的夹角,如果角度小于阈值,则连接面h与e.在邻接图中g与a,i连接,通过线性检测连接g和a.图9表示最终的邻接图G5.
Fig. 9 Discrete model group

图9 离散模型聚类

2.5 模型合并

本文采用Delaunay三角网合并三维模型[9]:(1)对建筑的角点进行合并,使其合并为一个平面;(2)对合并后的平面进行化简.图10表示模型合并的过程,图10(a)中虚线表示构造的约束Delaunay三角形,该三角形都有1个或者2个点在一个模型上,其他的点在另外的模型上,如果2个面属于同一个组,则合并约束三角网,图10(b)表示最后合并化简的结果.
Fig. 10 Polygon conflation

图10 底面合并

三维模型合并只考虑了平面方向,还需计算模型的垂直方向,即合并后三维模型的高度.假设合并 n 个模型,则合并后模型的高度 h 为式(3):
h = ∑ i = 1 n a i h i ∑ i = 1 n a i (3)
式中:ai代表模型 i 的底面面积,也为对应的权重;hi表示 i 模型的高度,由于三维模型的几何结构发生变化,需合并三维模型的纹理,并计算新的顶点纹理坐标.

3 分层次聚类算法实验与分析

本文选择北京市的一个典型小区作为实验区,图11(a)是通过房屋的轮廓和高度信息构造的2.5D三维模型;图11(b)是建筑对应的底面轮廓,其中曲线为道路;图11(c)是建筑顶面纹理.房屋轮廓和道路从矢量数据中获取,高度信息通过实地调绘获得,屋顶纹理从卫星影像中获取.在三维模型综合的过程中,一般可看见模型的屋顶纹理,侧面纹理会被其他模型遮挡,因此本文主要研究屋顶纹理.为了验证本文的方法,实验以Microsoft Windows 7系统为平台,用MATLAB 7.0实现主要的聚类和合并算法,采用OpenGL开发包实现三维模型群组的可视化.
Fig. 11 Experimental data

图11 研究区数据

实验步骤如下:(1)基于三维建筑模型的底面矢量数据(图11(b))构造约束Delaunay三角网,利用道路网分割三角网,分割后的三角网如图12(a)所示;(2)对三角网进行分析,删除第2类三角网,剩下的三角网表示可能的合并关系,图12(b)表示三角网分析后生成的邻接图;(3)对模型进行纹理分类,图12(c)是SOM智能分类算法,对每个子区域中的模型纹理分类的结果图,建筑上的编号表示其顶面纹理在子区域中的纹理类别;(4)通过纹理分析和模型聚类,进一步的处理邻接图,图12(d)表示对聚类的模型合并之后的效果图.
Fig. 12 Experimental result

图12 实验结果

经过实验处理,实验区聚类出41个群组,经检查这些群组内的各个建筑模型纹理,面积,高度,方向等特征较为一致,模型的排列也符合Gestalt准则,未形成聚类的多为相隔较远或者形状独特的地标建筑.与文献[7],[17]提出的单链接方法相比,本文通过基于建筑轮廓构造的三角网构建初始邻接图,避免了单链接方法的奇异问题,并且构造方法简单,可作为后续分类的候选链接.与文献[11]提出的基于认知的SVM聚类方法比较,本算法充分利用了三维模型的纹理特征,并保留了地标模型,更符合人的认知习惯.图13(a)和图13(b)分别表示合并后的不带纹理和带顶面纹理的三维可视化效果图.
Fig. 13 Model conflation

图13 模型合并

4 结论

本文提出了基于房屋轮廓及纹理的分层次聚类综合算法:基于三维模型的底面轮廓构造CDT,使用道路网分割CDT以保持城市形态学特征,并通过分析三角网构建初始邻接图;采用平均值和标准差指标,以SOM算法对纹理进行自动分类,使用局部区域分类的策略提高了分类准确度.本文发现邻接图可方便地应用Gestalt准则进行多层次三维模型聚类,结合建筑顶面纹理可提高聚类的效果.此外,在全局区域采用道路网分割模型群组,从而使模型保持城市形态学特征,在局部区域采用面积,方向等特征进行线性检测,可满足Gestalt准则,实现基于认知的聚类.从三维模型聚类可视化效果来看,其实验结果比较符合人的空间认知.另外,本方法由几个独立的算法模块分别完成相应处理步骤,可据实际情况对每个模块参数进行调整,具有较好的普适性.
但本文的方法今后还需进一步优化:该方法只考虑了屋顶纹理,未考虑侧面纹理,在视点比较近的地方,侧面纹理可能更重要,侧面纹理和屋顶纹理一起分类将是一个难点;本文对纹理的相似性度量比较简单,对纹理进行度量和分析还有待深化,以提升聚类效果;同时可考虑三维模型的语义信息,使用语义识别(包含三维模型的名称或者属性[23])以区分功能或外形相似的建筑,识别明显的地标模型,并将该类信息引入三维模型聚类综合的过程,以提高聚类效率与可视化效果.

The authors have declared that no competing interests exist.

参考文献

原文顺序 | 文献年度倒序 | 文中引用次数倒序
[1]
李德仁,赵中元,赵萍.城市规划三维决策支持系统设计与实现[J].武汉大学学报(信息科学版),2011,5(5):505-509,502.随着中国城市化进程的加快,城市建设面临日益复杂的空间关系与空间决策,特别是建设项目的规划决策不断影响着城市环境的协调性,基于二维平面图和效果图的规划决策已不能满足城市管理的需要。提出了建立真实现状三维环境和真实虚拟规划模型相结合的技术方法,研究了多种三维空间分析技术,开发了实用的系统,并投入工程化应用,取得了良好的效果。

[ Li D R, Zhao Z Y, Zhao P.Design and implementation of 3D decision support system for urban planning[J]. Geomatics and Information Science of Wuhan University, 2011,5(5):505-509,502. ]

[2]
Kada M.Aggregation of 3D buildings using a hybrid data approach[J]. Cartography and Geographic Information Science, 2011,38(2):153-160.With the increasing availability of large-area 3D city models, there is also a rising need for cartographic generalization algorithms to prepare this data so as to improve both the performance and the perceivability of their visualizations, especially in a mobile-mapping context. Most research has focused on the geometric simplification of single 3D buildings, which can only reach a certain degree of abstraction until building-specific geometric constraints are violated. To obtain further levels, an aggregation of building groups is necessary. The previously published building generalization approach based on cell decomposition is therefore extended to use morphological operations on a raster representation of the initially vectorial data. By repeatedly applying dilation and erosion, the space between buildings closes, which boosts the importance of the already available building blocks in these areas so that they are now included in the generalized model. Even though the results are encouraging from a technical point of view, it leads to a fundamental discussion of how the roofs of aggregated 3D buildings should appear.

DOI

[3]
Fan H, Meng L.A three-step approach of simplifying 3D buildings modeled by CityGML[J]. International Journal of Geographical Information Science, 2012,26(6):1091-1107.CityGML (City Geography Markup Language), the OGC (Open Open Geospatial Consortium) standard on three-dimensional (3D) city modeling, is widely used in an increasing number of applications, because it models a city with rich geometrical and semantic information. The underlying building model differentiates four consecutive levels of detail (LoDs). Nowadays, most city buildings are reconstructed in LoD3, while few landmarks in LoD4. For visualization or other purposes, buildings in LoD2 or LoD1 need to be derived from LoD3 models. But CityGML does not indicate methods for the automatic derivation of the different LoDs. This article presents an approach for deriving LoD2 buildings from LoD3 models which are essentially the exterior shells of buildings without opening objects. This approach treats different semantic components of a building separately with the aim to preserve the characteristics of ground plan, roof, and wall structures as far as possible. The process is composed of three steps: simplifying ...

DOI

[4]
Noskov A, Doytsher Y.Urban perspectives: A raster-based approach to 3D generalization of groups of buildings[C]. Proceedings of the GEO processing 2013, the Fifth International Conference on Advanced Geographic Information Systems, Applications, and Services, F, 2013.

[5]
Zhao J, Zhu Q, Du Z, et al.Mathematical morphology-based generalization of complex 3D building models incorporating semantic relationships[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2012,68:95-111.A complex 3D building model contains a detailed description of both its appearance and internal structure with authentic architectural components. Because of its high complexity and huge data volumes, using a less detailed representation for the distant visual application of such a model is preferable. However, most mesh simplification algorithms cannot preserve manmade features of such models, and the existing 3D generalization algorithms are mainly proposed for regular-shaped buildings. More importantly, neither method can consistently express geometry, topological relations, and semantics in multiple discrete Levels of Details (LoDs). This paper presents a novel mathematical morphology-based algorithm that generalizes the complex 3D building model in a unified manner using the following steps: (1) semantic relationships between components, which reflect structural connectivity in the building at a certain LoD, are defined and extracted; (2) semantically connected components are merged and trivial geometric features of the components are eliminated simultaneously, with semantics associated with components then updated according to the merging; and (3) post-process is carried out to further reduce the redundancy of facets. The semantic relationships extracted ensure the proper generalization of topological relations and semantics of building components, and mathematical morphological operations implemented in the algorithm are capable of handling closed two-manifold components of various shapes. Experiments on both complex 3D building models in the classical Chinese style and prismatic 3D city models prove the effectiveness of the proposed method.

DOI

[6]
Jiang S, Yang B, Sun X.Multi-resolution representation of 3D complex building models with features preservation[C]. Proceedings of the 2011 19th International Conference on Geoinformatics, IEEE, F, 2011.

[7]
Chang R, Butkiewicz T, Ziemkiewicz C, et al.Legible simplification of textured urban models[J]. IEEE Computer Graphics and Applications, 2008,28(3):27-36.CiteSeerX - Scientific documents that cite the following paper: Legible Simplification of Textured Urban Models

DOI

[8]
Lynch K.The image of the city[M]. Cambridge, MA: MIT press, 1960.

[9]
Yang L, Zhang L, Ma J, et al.Interactive visualization of multi-resolution urban building models considering spatial cognition[J]. International Journal of Geographical Information Science, 2011,25(1):5-24.Multi-resolution visualization of massive urban buildings is one of the most important components for a cyber city. Urban is a highly humanized system, and thus visualizing urban buildings needs to abide by people's habits of cognizing spatial relations between objects for their accurate and quick understanding of urban spatial information. This article proposes an approach to generalize and render urban building models in the context of Gestalt psychology and urban legibility. We introduce a new distance measurement method as the distance metric for the single-link clustering algorithm, which is used to group building footprints into clusters. Each cluster is merged based on the Delaunay triangulation and the polyline generalization algorithm. We then construct a hierarchical tree to store multi-resolution building models and implement interactive three-dimensional visualization of large-scale and high-density urban buildings. Experimental results indicate that the proposed methodology not only reduces the geometric complexity of urban models but also preserves urban legibility successfully and follow Gestalt principles.

DOI

[10]
Regnauld N.Contextual building typification in automated map generalization[J]. Algorithmica, 2001,30(2):312-333. Cartographic generalization aims to represent geographical information on a map whose specifications are different from those of the original database. Generalization often implies scale reduction, which generates legibility problems. To be readable at smaller scale, geographical objects often need to be enlarged, which generates problems of overlapping features or map congestion. To manage this problem with respect to buildings, we present a method of selection based on the typification principle that creates a result with fewer objects, but preserves the initial pattern of distribution. For this we use a graph of proximity on the building set, which is analysed and segmented with respect to various criteria, taken from gestalt theory. This analysis provides geographical information that is attached to each group of buildings such as the mean size of buildings, shape of the group, and density. This information is independent of scale. The information from the analysis stage is used to define methods to represent them at the target scale. The aim is to preserve the pattern as far as possible, preserve similarities and differences between the groups with regard to density, size and orientation of buildings. We present some results that have been obtained using the platform <i>Stratège</i> developed in the COGIT laboratory at the Institut Géographique National, Paris.

DOI

[11]
Zhang L, Deng H, Chen D, et al.A spatial cognition-based urban building clustering approach and its applications[J]. International Journal of Geographical Information Science, 2012,27(4):721-740.This article presents a spatial cognition analysis technique for automated urban building clustering based on urban morphology and Gestalt theory. The proximity graph is selected to present the urban mrphology. The proximity graph considers the local adjacency among buildings, providing a large degree of freedom in object displacement and aggregation. Then, three principles of Gestalt theories, proximity, similarity, and common directions, are considered to extract potential Gestalt building clusters. Next, the Gestalt features are further characterized with seven indicators, that is, area difference, height difference, similarity difference, orientation difference, linear arrangement difference, interval difference, and oblique degree of arrangement. A support vector machine (SVM)-based approach is employed to extract the Gestalt building clusters. This approach transforms the Gestalt cluster extraction into a supervised discrimination process. The method presents a generalized approach for clustering buildings of a given street block into groups, while maintaining the spatial pattern and adjacency of buildings during the displacement operation. In applications of urban building generalization and three-dimensional (3D) urban panoramic-like view, the method presented in this article adequately preserves the spatial patterns, distributions, and arrangements of urban buildings. Moreover, the final 3D panoramic-like views ensure the accurate appearance of important features and landscapes.

DOI

[12]
Guercke R, Götzelmann T, Brenner C, et al.Aggregation of LoD 1 building models as an optimization problem[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2011,66(2):209-222.<h2 class="secHeading" id="section_abstract">Abstract</h2><p id="sp000065">3D city models offered by digital map providers typically consist of several thousands or even millions of individual buildings. Those buildings are usually generated in an automated fashion from high resolution cadastral and remote sensing data and can be very detailed. However, not in every application such a high degree of detail is desirable. One way to remove complexity is to aggregate individual buildings, simplify the ground plan and assign an appropriate average building height. This task is computationally complex because it includes the combinatorial optimization problem of determining which subset of the original set of buildings should best be aggregated to meet the demands of an application.</p><p id="sp000070">In this article, we introduce approaches to express different aspects of the aggregation of LoD 1 building models in the form of Mixed Integer Programming (MIP) problems. The advantage of this approach is that for linear (and some quadratic) MIP problems, sophisticated software exists to find exact solutions (global optima) with reasonable effort. We also propose two different heuristic approaches based on the region growing strategy and evaluate their potential for optimization by comparing their performance to a MIP-based approach.</p>

DOI

[13]
Chang R, Wessel G, Kosara R, et al.Legible cities: Focus-dependent multi-resolution visualization of urban relationships[J]. IEEE Transactions on Visualization and Computer Graphics, 2007,13(6):1169-1175.Abstract Numerous systems have been developed to display large collections of data for urban contexts; however, most have focused on layering of single dimensions of data and manual calculations to understand relationships within the urban environment. Furthermore, these systems often limit the user芒s perspectives on the data, thereby diminishing the user芒s spatial understanding of the viewing region. In this paper, we introduce a highly interactive urban visualization tool that provides intuitive understanding of the urban data. Our system utilizes an aggregation method that combines buildings and city blocks into legible clusters, thus providing continuous levels of abstraction while preserving the user芒s mental model of the city. In conjunction with a 3D view of the urban model, a separate but integrated information visualization view displays multiple disparate dimensions of the urban data, allowing the user to understand the urban environment both spatially and cognitively in one glance. For our evaluation, expert users from various backgrounds viewed a real city model with census data and confirmed that our system allowed them to gain more intuitive and deeper understanding of the urban model from different perspectives and levels of abstraction than existing commercial urban visualization systems.

DOI PMID

[14]
Glander T, Döllner J.Abstract representations for interactive visualization of virtual 3D city models[J]. Computers, Environment and Urban Systems, 2009,33(5):375-387.Virtual 3D city models increasingly cover whole city areas; hence, the perception of complex urban structures becomes increasingly difficult. Using abstract visualization, complexity of these models can be hidden where its visibility is unnecessary, while important features are maintained and highlighted for better comprehension and communication. We present a technique to automatically generalize a given virtual 3D city model consisting of building models, an infrastructure network and optional land coverage data; this technique creates several representations of increasing levels of abstraction. Using the infrastructure network, our technique groups building models and replaces them with cell blocks, while preserving local landmarks. By computing a landmark hierarchy, we reduce the set of initial landmarks in a spatially balanced manner for use in higher levels of abstraction. In four application examples, we demonstrate smooth visualization of transitions between precomputed representations; dynamic landmark highlighting according to virtual camera distance; an implementation of a cognitively enhanced route representation, and generalization lenses to combine precomputed representations in focus+context visualization.

DOI

[15]
Glander T, Döllner J.Automated cell based generalization of virtual 3D city models with dynamic landmark highlighting[C]. Proceedings of the 11th ICA Workshop on Generalization and Multiple Representation, Monpellier, France, F, 2008.

[16]
Zhang M, Zhang L, Mathiopoulos P T, et al.A geometry and texture coupled flexible generalization of urban building models[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2012,70(6):1-14.In the past, numerous research efforts have focused on generalization of city building models. However, a generic procedure for creating flexible generalization results supporting the fast and efficient update of original building models with various complexities is still an open problem. Moreover, building clusters created in previously published generalization methods are not flexible enough to meet the various requirements for both legible and realistic visualization. Motivated by these observations, this paper proposes a new method for generating a flexible generalization outcome which enables convenient updating of original building models. It also proposes a flexible preprocessing of this generalized information to render a legible and realistic urban scene. This is accomplished by introducing a novel component structure, termed as FEdge, particularly designed for efficiently managing the geometry and texture information in building cluster instances (both original building models and building clusters) during the generalization, visualization and updating processes. Furthermore, a multiple representation structure, referred to as Evolved Buffer-Tree (EBT), is also introduced. The purpose of the EBT is to organize building cluster instances and to employ more flexible LODs for both legible and realistic visualization of urban scenes. FEdge has an intuitive planar shape which can be effectively used in representing rough 3D facade composed by detailed continuous meshes. Each FEdge is given a unique identifier, referred to as FEdge Index. In the proposed generalization scheme, firstly each original building model treated as a building cluster instance is abstracted and presented as FEdge Indices. These FEdge Indices are then used for producing generalized building cluster instances in the EBT portably, and to support convenient model updating and flexible preprocessing of the generalization results for renderable building cluster instances. Secondly, to achieve a legible and realistic visualization of urban scene, the EBT is flexibly assigned diverse LODs maintaining more important legible information than LODs defined in CityGML for 3D building models. To make the generalization more accurate by considering the city roads and districts, an algorithm for automatic road analysis is applied in our clustering and combination. Numerous experiments considering the geometrical and textural complexity of common urban building models, as well as a typical case study of complex city scene with a large number of building models, verify the effectiveness of our generalization method and the dynamic visualization of the generalized urban models.

DOI

[17]
Yan H, Weibel R, Yang B.A multi-parameter approach to automated building grouping and generalization[J]. GeoInformatica, 2008,12(1):73-89.<a name="Abs1"></a>This paper presents an approach to automated building grouping and generalization. Three principles of Gestalt theories, i.e. proximity, similarity, and common directions, are employed as guidelines, and six parameters, i.e. minimum distance, area of visible scope, area ratio, edge number ratio, smallest minimum bounding rectangle (SMBR), directional Voronoi diagram (DVD), are selected to describe spatial patterns, distributions and relations of buildings. Based on these principles and parameters, an approach to building grouping and generalization is developed. First, buildings are triangulated based on Delaunay triangulation rules, by which topological adjacency relations between buildings are obtained and the six parameters are calculated and recorded. Every two topologically adjacent buildings form a potential group. Three criteria from previous experience and Gestalt principles are employed to tell whether a 2-building group is &#8216;strong,&#8217; &#8216;average&#8217; or &#8216;weak.&#8217; The &#8216;weak&#8217; groups are deleted from the group array. Secondly, the retained groups with common buildings are organized to form intermediate groups according to their relations. After this step, the intermediate groups with common buildings are aggregated or separated and the final groups are formed. Finally, appropriate operators/algorithms are selected for each group and the generalized buildings are achieved. This approach is fully automatic. As our experiments show, it can be used primarily in the generalization of buildings arranged in blocks.

DOI

[18]
Mao B, Ban Y, Harrie L.A multiple representation data structure for dynamic visualisation of generalised 3D city models[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2011,66(2):198-208.<h2 class="secHeading" id="section_abstract">Abstract</h2><p id="sp000175">In this paper, a novel multiple representation data structure for dynamic visualisation of 3D city models, called CityTree, is proposed. To create a CityTree, the ground plans of the buildings are generated and simplified. Then, the buildings are divided into clusters by the road network and one CityTree is created for each cluster. The leaf nodes of the CityTree represent the original 3D objects of each building, and the intermediate nodes represent groups of close buildings. By utilising CityTree, it is possible to have dynamic zoom functionality in real time. The CityTree methodology is implemented in a framework where the original city model is stored in CityGML and the CityTree is stored as X3D scenes. A case study confirms the applicability of the CityTree for dynamic visualisation of 3D city models.</p>

DOI

[19]
Allouche M K, Moulin B.Amalgamation in cartographic generalization using Kohonen's feature nets[J]. International Journal of Geographical Information Science, 2005,19(8-9):899-914.Empirical observations of the way cartographers deal with generalization problems lead to the hypothesis that they first detect patterns of anomalies in the cartographic data set and then eliminate anomalies by transforming the data. Automatically identifying patterns of anomalies on the map is a difficult task when using GIS functions or traditional algorithmic approaches. Techniques based on the use of neural networks have been widely used in artificial intelligence in order to solve pattern‐recognition problems. In this paper, we explore how Kohonen‐type neural networks can be used to deal with map generalization applications in which the main problem is to identify high‐density regions that include cartographic elements of the same type. We also propose an algorithm to replace cartographic elements located in a region by its surrounding polygon. The use of this type of neural network permitted us to generate different levels of grouping according to the chosen zoom‐scale on the map. These levels correspond to a multiple representation of the generalized cartographic elements. As an illustration, we apply our approach to the automatic replacement of a group of houses represented as a set of very close points in the original data set, by a polygon representing the corresponding urban area in the generalized map.

DOI

[20]
杨晓敏,严斌宇,吴炜,等.基于图像色彩和纹理的SOM聚类和检索方法[J].四川大学学报(自然科学版),2010,47(3):525-529.作者根据图像的全局色彩和区域纹理信息,使用自组织映射神经网络的方法对图像内容进行聚类和 检索.全局色彩在HSI空间使用区域累加的方法,避免了维数过大的问题.用区域纹理描述的方法解决了使用单一色彩带来的不准确性;自组织映射网络所特有的 特征选择和无监督学习等特性,实现了对视觉相似图像的聚类.

DOI

[ Yang X M, Yan B Y, Wu W, et al.Clustering and retrieval using self-organized mapping networks based on image color and texture[J]. Journal of Sichuan University(Natural Science Edition), 2010,47(3):525-529. ]

[21]
Mao B, Harrie L, Ban Y.Detection and typification of linear structures for dynamic visualization of 3D city models[J]. Computers, Environment and Urban Systems, 2012,36(3):233-244.Cluttering is a fundamental problem in 3D city model visualization. In this paper, a novel method for removing cluttering by typification of linear building groups is proposed. This method works in static as well as dynamic visualization of 3D city models. The method starts by converting building models in higher Levels of Details (LoDs) into LoD1 with ground plan and height. Then the Minimum Spanning Tree (MST) is generated according to the distance between the building ground plans. Based on the MST, linear building groups are detected for typification. The typification level of a building group is determined by its distance to the viewpoint as well as its viewing angle. Next, the selected buildings are removed and the remaining ones are adjusted in each group separately. To preserve the building features and their spatial distribution, Attributed Relational Graph (ARG) and Nested Earth Mover鈥檚 Distance (NEMD) are used to evaluate the difference between the original building objects and the generalized ones. The experimental results indicate that our method can reduce the number of buildings while preserving the visual similarity of the urban areas.

DOI

[22]
Li Z, Yan H, Ai T, et al.Automated building generalization based on urban morphology and Gestalt theory[J]. International Journal of Geographical Information Science, 2004,18(5):513-534.Building generalization is a difficult operation due to the complexity of the spatial distribution of buildings and for reasons of spatial recognition. In this study, building generalization is decomposed into two steps, i.e. building grouping and generalization execution. The neighbourhood model in urban morphology provides global constraints for guiding the global partitioning of building sets on the whole map by means of roads and rivers, by which enclaves, blocks, superblocks or neighbourhoods are formed; whereas the local constraints from Gestalt principles provide criteria for the further grouping of enclaves, blocks, superblocks and/or neighbourhoods. In the grouping process, graph theory, Delaunay triangulation and the Voronoi diagram are employed as supporting techniques. After grouping, some useful information, such as the sum of the building's area, the mean separation and the standard deviation of the separation of buildings, is attached to each group. By means of the attached information, an appropriate operation is selected to generalize the corresponding groups. Indeed, the methodology described brings together a number of well-developed theories/techniques, including graph theory, Delaunay triangulation, the Voronoi diagram, urban morphology and Gestalt theory, in such a way that multiscale products can be derived.

DOI

[23]
Al-Bakri M, Fairbairn D.Assessing similarity matching for possible integration of feature classifications of geospatial data from official and informal sources[J]. International Journal of Geographical Information Science, 2012,26(8):1437-1456.ABSTRACT One difficulty in integrating geospatial data sets from different sources is variation in feature classification and semantic content of the data. One step towards achieving beneficial semantic interoperability is to assess the semantic similarity among objects that are categorised within data sets. This article focuses on measuring semantic and structural similarities between categories of formal data, such as Ordnance Survey (OS) cartographic data, and volunteered geographic information (VGI), such as that sourced from OpenStreetMap (OSM), with the intention of assessing possible integration. The model involves &lsquo;tokenisation&rsquo; to search for common roots of words, and the feature classifications have been modelled as an XML schema labelled rooted tree for hierarchical analysis. The semantic similarity was measured using the WordNet::Similarity package, while the structural similarities between sub-trees of the source and target schemas have also been considered. Along with dictionary and structural matching, the data type of the category itself is a comparison variable. The overall similarity is based on a weighted combination of these three measures. The results reveal that the use of a generic similarity matching system leads to poor agreement between the semantics of OS and OSM data sets. It is concluded that a more rigorous peer-to-peer assessment of VGI data, increasing numbers and transparency of contributors, the initiation of more programs of quality testing and the development of more directed ontologies can improve spatial data integration.

DOI

Options
文章导航

/

〈 〉