面向三维建筑的过程建模语法设计、推导和逆向预测

Procedural modeling enjoys wide applications in 3D content creation, urban modeling and image data generation. Existing procedural modeling system requires tedious modeling, a long learning curve, and often encounters difficulty when modeling the variations. This project proposes new methods, which feature in the semantic modeling process, constraint template-based model regularization, grammar induction and inverse procedural rule prediction. The key scientific issues to be tackled include semantic-oriented procedural modeling, similarity metrics for the consolidation of procedural rules, and inverse procedural rule prediction based on point cloud. The innovativeness of the project is as follows: introducing structural shapes, selection expressions and constraint templates to facilitate more natural modeling process; exploiting synthetic data to learn the similarity between geometry and measuring the similarity between procedural rules based on Hungarian algorithms; and using depth learning algorithms to predict the procedural rules that generate the model of point clouds, and the next modeling rules of a modeling geometry. This project helps to promote the development of more natural and intelligent procedural modeling and will directly serve the digital content production industry.

过程建模在三维内容创作、城市建模和图像数据生成等诸多方面有着广泛的应用。本项目基于现有过程建模系统过于繁琐、学习曲线长和多样性建模难等特点提出面向语义的过程建模算法，基于约束模板的规则化建模，服务于多样性合成的语法规则推导和逆向语法规则预测。拟解决的关键科学问题包括：面向语义的过程建模设计、语法规则的相似性分析和合并、基于点云的逆向语法规则预测。项目的创新性在于：引入结构形状、选择表达式和约束模板来实现更自然的过程建模方式；使用合成数据学习几何模型相似性并基于匈牙利算法分析语法规则的相似性；使用深度学习算法预测点云对应的语法规则和下一条建模规则。本项目的研究有助于促进更加自然、智能的过程建模方式的发展，并服务于数字内容制作行业。

过程建模在虚拟内容创作、建筑和树木建模、城市规划设计等不同领域有着广阔的应用前景和重要的价值。然而，如何有效的设计过程模型使其更符合人的理解认知，并能将其与机器学习算法进行结合，实现对大量建模实例的分析是目前的一个研究难点和重点。本项目围绕过程模型的语法设计、逆向预测和推导进行研究，并着重解决了一下问题：1）服务于三维建筑建模的有效、自然的过程模型设计；2）基于实例模型的自动化逆向过程模型重建；3）基于过程模型的语法推导与泛化；4）针对三维模型的实例化分析技术。通过研究以上问题，本项目取得了多项创新性研究成果，其中包括CCF-A类会议与期刊论文四篇，CCF-B类期刊论文一篇，并获专利授权一项，培养研究生三名。

内容获取失败，请点击重试