基于几何理解的交互式实例建模技术研究

To achieve the creative reuse of digital geometry models, this project explores techniques and methodologies of the interactive modeling by example based on geometric structure discovery, focused on three main aspects: the methodes of geometric structure discovery, the sketch-based interactive mode for structure modeling, and the semantic-driven model evolving mechanism. Firstly, it would adopt the generalized various granularity definition of structural unit and self-adaptive fitting mechanism to code and extract semantic structure of the model. Secondly, context-sensitive viewpoint selection and line drawing strategy would be combined with stroke correspondences algorithm based on decision optimization to enhance the controllability of interactive structure modeling process. Thirdly, the combinatorial optimization and constraint propagation mechanism of hierarchical structure features are performed to support the preservation of the structural characteristics and the adaptability to complex structure during model evolution. Furthermore, a integrated methodology for the whole process of interactive modeling by example would be established, and a service-oriented application platform for the creation and reuse of model resources in the 3D animation or game modeling would be developed. The research results will not only meet the demand of application diversification and the various types of resource-reusing in design fields such as industrial design and digital entertainment, but also activate and contribute some aspects to the growing of semantic digital geometry processing and applications.

本项目从模型结构的几何理解方法、结构建模的草绘交互方式和语义驱动的模型演变机制等三个主要方面，探索基于几何理解的交互式实例建模技术及方法，以支持数字几何模型的创作性重用。拟采用广义变粒度几何结构单元定义及其自适应拟合机制来实现模型结构的语义化几何理解；采取内容敏感的视角选择和线绘制与笔画匹配的决策优化相结合策略来增强结构建模过程的交互可控性；运用层次化结构特征组合优化与约束传播机制来支持模型演变过程的特征保持及其对复杂模型结构的适应性；进而,建立相对完整的实例建模方法体系并构建面向三维动漫游戏模型创作与资源重用服务应用平台。项目成果将不仅能满足工业造型、数字娱乐等数字设计领域中设计模式及其资源重用的需求，而且可拓宽数字几何处理语义化生成与应用发展方向的研究思路和技术途径。

本项目顺应新工业革命“大众创造”的发展趋势，以支持大众化创意建模方法和软件技术为研究目标，对基于几何理解的交互式实例建模技术进行了系统而深入的探索，通过将数字几何模型的分类组织、语义化分析和结构演变进行有机结合，利用实例的编辑、变异、替换和组合等多种模式实现新三维模型的构造。项目所取得的成果涵盖以三维内容辅助创作为应用切入点的数字几何语义化分析、草绘交互以及实例化生成等三大方面：一是针对数字几何模型的内容标注和部件提取，采用在线学习、增量更新和用户驱动等技术实现了三维模型集的渐进式分割和迭代式分类，解决了模型语义化分析方法的模型集更新和用户交互指导问题。二是针对面向几何建模的草绘交互，提出了基于在线度量学习的草图标注和基于人工免疫的草图数据集扩展方法，提高了草图内容识别的性能；同时，提出了基于类别判别性和粒子群优化的视角选择、基于条件随机场的特征线提取及决策优化的特征线匹配等方法，解决了实例建模中用户驱动的信息呈现和交互体验等问题。三是研究了涵盖单个模型到大规模场景的实例化生成技术，提出了基于重复结构发现和最大期望优化的模型合成、结合图像指导和模型驱动的场景生成以及结合三维扫描、LED标记跟踪和激光测距的三维重建方法，解决了数字几何模型的变粒度实例结构组合生成、复杂结构约束传播、高维空间优化和物理对象的数字化重构等问题。项目根据已有技术成果，构建了完整的基于几何理解的交互式实例建模技术和方法体系，并设计开发了相应的实例建模自助式平台与软件工具，在创意建模、工业生产、城市规划等多个领域相关模型创作的实验验证了系统的有效性。