基于IFC的建筑信息模型(BIM)语义检索技术研究

Semantic-based 3D model retrieval is the current research hot topic, which should not only consider the geometric features of 3D models, but also focus on the domain-specific semantic features. With the rapid popularity of Building Information Modeling (BIM) technologies in AEC field, BIM model resources such as building product libraries are growing rapidly on the Web. Meanwhile, how to efficiently search such BIM models becomes increasingly urgent. However, the conventional methods for 3D model retrieval are mostly based on keyword matching or geometric shape matching, which cannot provide effective support for design reuse applications due to the lack of extraction, representation and matching of semantic information in AEC field. This proposal presents a new methodology for semantic-based BIM model retrieval, especially for the Web building product libraries. We also develop a vertical search engine for BIM model retrieval, which will help users to quickly find the 3D models related to design reuse. Our work consists of several parts as follows. Firstly, based on Industry Foundation Classes (IFC) schema that is a major data exchange standard for BIM, we construct a domain-specific ontology for encoding the domain knowledge and extend the ontology through machine learning techniques. Secondly, an annotation and indexing method for BIM models is presented based on the geometric and semantic features. Thirdly, we propose a semantic similarity measure for BIM models based on Tversky measure and information content techniques. Finally, an ontology query expansion method is presented based on SPARQL and LCA-based concept tree pruning techniques.

三维模型语义检索是当前研究热点，语义检索不仅要考虑检索对象的几何形状特征，还应关注应用领域的语义特征。近年来，在建筑工程领域随着BIM（Building Information Modeling）技术的快速发展和普及，互联网上三维BIM模型正迅猛增长，如何有效地检索这些模型资源变得日益迫切。传统的三维模型检索大多基于关键字匹配、几何形状相似匹配方法，而缺乏对建筑工程领域语义信息的提取、表示和匹配，不能很好地满足用户查询，难以有效地支持设计复用。本课题针对互联网上建筑产品库开展BIM模型语义检索技术研究，并开发一个针对BIM模型的垂直搜索引擎原型，有助于建筑设计的快速查找和有效复用。主要研究内容包括：基于IFC标准的BIM领域本体构建和学习，基于几何与语义混合特征的BIM模型自动标引，基于Tversky模型与信息量的BIM模型语义相似性度量，基于SPARQL与LCA剪枝的本体扩展查询方法。

本课题针对互联网上建筑产品库开展BIM模型语义检索技术研究，并开发了一个针对BIM模型的垂直搜索引擎原型，有助于建筑设计的快速查找和有效复用。主要成果包括：（1）提出一种基于内容的IFC文件压缩算法，在此基础上实现了IFC模型快速离散、传输和加载方法，大大优化了IFC文件内容及其在Web端模型显示的效率；（2）提出一种三维模型的可视内部体积度量方法，并将该度量方法应用到三维模型的关节语义特征提取、模型分割、形状匹配与检索；（3）提出一种基于BoSCC (Bag of Spatial Context Correlations) 的三维模型形状表征方法，增强了传统基于Bag of Words (BoW)方法的空间信息表征能力，该方法被应用到三维模型形状分类、全局形状检索和局部形状检索中；（4）提出一种基于三维模型序列视图表示的三维模型形状表征方法，该方法考虑到了序列视图之间存在的顺序关系和不同视角下的视图对三维模型表示的不同重要程度来增强三维模型的特征表示；（5）提出一种基于三维模型点云表示的形状表征学习方法，该方法关注于获取点云局部邻域内的细粒度上下文信息，并引入注意力（Attention）机制来突出邻域内不同尺度区域的重要程度；（6）提出一种共同表征三维模型和自然语言两种不同模态并可实现两者间的跨模态检索的方法，该方法可用于扩展检索方式，使检索结果形式更加多样化。项目期间共发表或接收论文13篇，其中SCI索引论文9篇，国际会议论文4篇，包括CCF A类论文6篇。此外，1篇论文获得国际会议ICCCBE 2016的Best Student Presentation奖。