基于CityGML的明清古建筑三维语义模型扩展及转换研究

针对现有的大量古建筑三维模型缺乏语义信息,不利于精细化和智能化管理等问题,对城市地理标记语言（City Geography Markup Language, CityGML）定义的建筑物模型进行了语义扩展,提出了一种从古建筑几何模型自动识别语义对象并转化为CityGML扩展模型的方法。首先,总结我国明清古建筑构件的共有特征,并使用CityGML的应用领域扩展机制增加斗拱、梁、柱、台基等古建筑通用对象的几何和语义表达。其次,利用古建筑Mesh模型各面片的法向量、坐标范围等特征和自定义规则提出了一套古建筑屋顶、墙面、立柱、门窗等语义对象自动识别方法。最后,依据CityGML扩展模型设计规范,从识别对象自动构建出几何语义一体化表达的古建筑三维模型。采用2个不同风格的古建筑模型进行语义对象提取和CityGML模型生成实验,使用本文提出的Mesh模型到CityGML LoD3扩展语义模型转换方法能够自动提取屋顶、墙面、台基等9类语义对象,每个Mesh模型能转换并生成400多个包含语义信息的平面多边形,95%以上的面片能正确识别。结果表明本文设计的CityGML扩展模型能够有效支持古建筑模型典型构件的显式语义表达,并且识别规则和转换方法能够支持Mesh模型大多数平面对象的提取,有利于CityGML模型的自动化生成和精细化管理。

3D Semantic Model Extension and Transformation for Ancient Architecture of Ming and Qing Dynasties based on CityGML

In recent years, a large amount of 3D building models of Chinese ancient architecture have been created to preserve and protect cultural heritages by different organizations and persons. However, most of these existing 3D models are lack of semantic information, which are predominantly used for 3D visualization, and they cannot meet the requirements of fine management and other intelligent applications. To address this problem, the standard building model defined by City Geography Markup Language (CityGML) is extended based on the features of Chinese ancient architecture. Moreover, a novel 3D semantic modeling approach is proposed to automatically identify semantic surfaces from mesh models and generate extended CityGML model in this paper. Firstly, we summarize the common features of Chinese ancient architecture components in Ming and Qing dynasties. We develop a CityGML extension to explicitly represent the geometric and semantic information of typical components of ancient architectures using CityGML Application Domain Extensions (ADE). Bracket, beam, column, base, and other new objects are added into this designed CityGML building model. Secondly, a model transformation algorithm is proposed to generate extended CityGML building models from mesh models, in which face normal and coordinate range of each triangle face from mesh model are calculated, together with other defined shape and position rules for different components. The algorithm is able to automatically recognize different semantic surfaces from roof, wall, column, and other objects of an ancient building. Thirdly, all these extracted triangular faces are further refined via topological validation and merged into polygon geometries, so as to accurately represent an ancient building according to geometric and semantic principles of CityGML. Two public mesh models of ancient buildings with different construction structures are selected to automatically extract semantic objects and generate corresponding extended CityGML models using the proposed method. Nine kinds of semantic objects such as RoofSurface, WallSurface, and Base are successfully recognized. More than 400 planar polygons with semantic information for each CityGML building model are generated. The percentage of correct face recognition is as high as 95%. Experimental results indicate that the extended CityGML model can effectively support the explicit semantic representation of typical components for ancient architecture. Most of planar faces from these mesh models can be automatically extracted and converted into extended CityGML semantic objects using the explored geometric rules and transformation approach. Therefore, the presented algorithm is beneficial for 3D semantic modeling of Chinese ancient architectures in an automated manner. It is useful for the fine management of ancient building models as well.