【作者】 胡春；

【导师】 田金文；

【作者基本信息】 华中科技大学 ， 模式识别与智能系统， 2004， 硕士

【摘要】 随着GIS的迅速发展,对城市环境中各种三维信息的表达与处理变得日益迫切,需要建立富有管理机能和操作机制的三维城市模型系统来对大批量三维数据进行管理和组织。本文以建立三维城市模型系统为应用背景,对高程图中的建筑物全自动提取、建筑物三维模型的表达以及城区三维景观的重建做了系统的分析与深入的研究,通过大量的实验对相关理论和算法进行了验证。首先,利用深度图像分割的原理对高程图中的建筑物实现全自动提取。考虑到建筑物信息包含在原始高程与地形表面近似估计的差值图像中,利用灰度数学形态学的方法可对原始高程图进行背景估计和消除。随地形起伏动态选择阈值,实现自适应阈值法的全局分割。依据三维表面法向量分布的有序或杂乱,完成对建筑物或植被区域的识别。用双向边界算子跟踪特征的边界,以树形结构表示特征边界之间的相互关系,实现了网格数据场中特征区域的边界跟踪。引入Freeman链码法和基于聚合的最小均方误差多边形逼近来对分割出的特征区域进行边界的矢量表达。其次,考虑到三维城市景观实体的特点,结合三维城市模型中景观对象的表达需求分析了基于几何元素、基于拓扑关系和基于多面体模块的建筑物建模方法及各自特点。根据建筑物的结构和对称性提取几何信息,用参数变量加以描述,给出了几种典型建筑物的参数模型和相应的数据表达。最后,实现了基于DEM叠加航空影像的城区三维景观重建。根据被提取出的建筑物及其参数模型,尝试运用建筑物与地形的分层建模方法重建城区三维场景,并分析了该方法带来的建筑物与地形的融合问题及相应的解决策略。针对依次连接相邻网格点为多边形的绘制方式带来的建筑物墙面倾斜、建筑物与地形边界混淆等问题,根据建筑物地形轮廓线的几何特征,提出了一种在规则网格高程场中提取建筑物并重建三维场景的算法。更多还原

【Abstract】 With the rapid development in GIS, the representation and processing of three-dimensional information become greatly in demand. A three-dimensional city model with the ability of management and manipulation should be established in order to manage and organize 3d data in large quantity. This thesis explores to validate building extraction from altitude data, the representation of building models and the reconstruction of 3d city landscape.Firstly, the theories of range image segmentation are applied to automatic building extraction from digital elevation models. The information about the buildings is contained in the difference between original DEM and an approximation of topographic surface without buildings. Therefore we use mathematical grey scale morphology to realize the background estimation and elimination. The threshold for segmentation can be dynamic motivated by altitude information of terrain surface and derived from prior knowledge about the buildings. A building is defined as a set of planar patches in the same 3d region, which enables us to define criteria for buildings from our 3d data. Building and vegetation classification relies on the study of distribution of the normal to 3d local surface.Two edge operators are used to track the border of feature areas. The relationships of feature borders are represented by a tree structure. Freeman chain codes and a merging algorithm are applied in order to convert feature areas from raster to vector.Secondly, methods of data modeling based on geometry primitives, blocks and topology are analyzed. According to building structures and geometry constraints, several typical simple building parametric models are given.Finally, a city landscape is built up based on DEM and aerial imagery. On the basis of building extraction, 3d city landscape can be built up by reconstructing buildings and terrain separately. But this also brings the matching problem of buildings and terrain, which should <WP=5>be resolved by model transform. The general rendering method in digital elevation model including building altitude information brings the problems of sloping building walls and the mixture of buildings and terrain. An approach of extracting buildings from digital elevation model data by utilizing the great altitude difference between building vertices and terrain vertices, and the geometry characteristics of buildings is developed.更多还原