基于上下文感知的部件组装三维建模

项目名称： 基于上下文感知的部件组装三维建模

项目编号： No.61202223

项目类型： 青年科学基金项目

立项/批准年度： 2013

项目学科： 计算机科学学科

项目作者： 谢晓华

作者单位： 中国科学院深圳先进技术研究院

项目金额： 24万元

中文摘要： 基于部件组装的三维建模方法主要根据用户的交互输入从模型库中检索用户感兴趣的模型并提取相关部件组装成新的三维模型。该建模方式实现对已有三维模型资源的复用而无须从零开始建模，符合大众的创造规律，因此近年受到相关研究人员的高度重视。然而在建模过程中，用户的输入往往具有模糊性和随意性，甚至用户对建模对象的几何形式尚未形成完整清晰的概念，需要得到即时启发。针对这些问题，本项目拟以家具、工艺品等人造三维目标为建模对象，研究在部件组装建模中引入上下文信息（如几何纹理一致性、形状协调性、接口兼容性等部件间潜在的关联及建模实际环境约束等），从而更好地理解用户的意图并为用户提供合理的建议和启发。研究的关键问题包括对三维模型多尺度视觉属性的分开提取与匹配以及基于上下文感知的三维模型检索。本项目的研究目标是丰富对三维模型的理解与分析理论，探索上下文信息在三维建模中的应用，进一步完善基于部件组装的三维建模方法。

中文关键词： 上下文；三维建模；图像理解；语义分割；形状检索

英文摘要： The assembly-based 3D modeling method is to retrieve some interesting 3D models from a database according to user's interactive input, and then extract the relevant shape components from these retrieved models to assemble into one or more than one desired 3D models. The assembly-based 3D modeling method recycles preexisting 3D model source rather than modeling from scratch, which is in line with the rule of masses creation therefore has been highly valued by correlative researchers. In the modeling process, however, the user's input is always with ambiguousness and contingency even that, the user has no clear idea about the geometrical form of the designing object. In this case, the user should be immediately inspired from the system. Accordingly, taking man-made 3D objects such as as furnitures and handicrafts as instances, in this project we attempt to introduce the contextual information (e.g., the consistency of geometry textures, the coordination of shapes, and the compatibility of connectors between 3D parts, and constraints of practical modeling environmental, and so on) to the assembly-based 3D modeling process. With the help of contexts, the system can understand the intention of customer better and provide the user the reasonable suggestions and inspirations. The key issues of this project include the

英文关键词： context；3d modeling；image understanding；semantic segmentation；shape retrieval