项目名称: 数据驱动的非刚体几何模型注册新方法研究
项目编号: No.61502453
项目类型: 青年科学基金项目
立项/批准年度: 2016
项目学科: 计算机科学学科
项目作者: 高林
作者单位: 中国科学院计算技术研究所
项目金额: 21万元
中文摘要: 三维模型注册是数字几何处理的关键技术之一,一直是图形学研究的重点和热点。其中刚体注册技术已取得极大的进展,然而非刚体注册依然是难点,尤其针对具有大尺度非刚体变形的情况。这是因为非刚体注册的解空间更大,同时是非线性的。如何注册由这些深度相机扫描得到的具有大尺度非刚体变形的点云成为亟待解决的难题。同时随着扫描技术的发展,一些消费级深度相机涌现出来。其具有造价低体积小的优点,并且更加容易实时地捕获到具有大尺度变形的点云。因此非刚体注册技术有着广泛的应用前景。.基于这一背景,本课题拟研究数据驱动的非刚体几何模型注册新方法。通过对变形模型库进行特征提取与分析,并将其作为非刚体注册的先验信息来缩小解空间。从而可以注册具有大尺度非刚体变形的低质点云。为此,本课题将研究可以表达模型库中大尺度变形的几何特征,研究特征集合的主分量分析与传播;研究基于约束的非刚体几何合成方法;研究基于先验的非刚体几何注册方法。
中文关键词: 点云注册;数据驱动;网格表示;旋转不变
英文摘要: Shape registration is one of the key technologies of digital geometry processing that is the research hotspot in computer graphics. Rigid registration techniques have made great progress, but non-rigid registration is still difficult, especially for the case with large-scale non-rigid deformation. This is because the solution space of non-rigid registration is larger than rigid registration, and it is not linear. How to register a point cloud with a large-scale non-rigid deformation form the depth camera become an urgent problem to be solved. Simultaneously with the development of scanning technology, some consumer depth cameras emerge which has the advantage of low cost, small size. And it is easier to capture the point cloud with large-scale deformation. Therefore, non-rigid registration technology has a broad application prospects..Based on this background, this study intends to develop a novel data-driven registration method for the point cloud with large scale deformation. Through feature extraction and analysis, we treat the collection of features as a priori information to reduce the solution space. Thus the low-quality point cloud with large-scale non-rigid deformation could be registered. This paper will propose a novel geometric characteristics to express the large-scale deformation and research principal component analysis; the feature collection propagation; synthesis method with constraint and data driven based non-rigid registration method.
英文关键词: registration;data-driven;mesh representation;rotation invariant