散乱点集多域非均质物体内部分界面提取方法研究

This project mainly focuses on the multi-domain-material objects expressed by scattered point data fields, and the content of the main researches consist of the extraction of directed skeleton tree, the structure of discrete surface primitives and the expression of complex surface. Furthermore, we explore the difficult problem in the recognition of internal material interface, and then we obtain a novel interface extraction method. Including the following details: Firstly, according to the structure of multi-domain-material heterogeneity, we establish a clustering by fast search algorithm and a multivariate transfer model to improve the degree of automation of internal material classification markings on multi-domain-material objects. Secondly, to provide a powerful tool for accurately extracting topological structures among each material of multi-domain-material objects, we explore the spatial variation rule of the internal material of multi-domain-material objects and construct a directed skeleton tree model with global topological features. Finally, in order to improve the accuracy of multi-domain-material interface representation, the structural rules of complex surface are studied based on point set, and a strategy based on Reeb graph and minimum enclosing activity contours is proposed to extract a single-material discrete surface primitives of multi-domain-material. This achievement not only provide a new way to recognize the complex interface of multi-domain-material object in scattered data field, but also enrich the theory and methodology of pattern recognition based on three-dimensional point set, and further provide a fundamental support for the application of medical analysis, geological exploration, marine water research, space navigation and so on.

该课题以散乱点集数据场表达的多域物体为研究对象，以有向骨架树的提取、离散曲面基的构造和复杂曲面的表达等为主要研究内容，探索其内部材质分界面这一复杂曲面的识别难题，拟获得新的分界面提取方法。具体包括：⑴顾及多域物体内部结构的异质性，建立基于密度峰值的快速聚类算法和多变量传递模型，提高对多域物体内部材质分类标记的自动化程度；⑵探索多域物体内部材质的空间变异规律，构建一种具有全局拓扑特征的有向骨架模型，为准确提取多域物体材质之间的拓扑结构提供有力工具；⑶研究点集复杂曲面的构造规律，给出基于Reeb图及最小外接活动轮廓提取多域物体单材质离散曲面基的策略，有效提高多域物体分界曲面表达的准确性。其成果可为散乱数据场多域物体复杂分界面的识别提供一种新的解决途径，丰富计算机视觉智能的理论与方法体系，进而为医学分析、地质勘探、海洋水体研究、空间导航等领域的应用提供基础支撑。

该课题以散乱点集数据场为核心，以多域物体为研究对象，以“多域物体分界曲面提取与表达方法体系的建立”为研究目标，探索其内部材质分界面这一复杂曲面的提取难题。取得的主要研究成果可概括为：(1)建立了基于有向性的全局空间拓扑结构；(2)设计了多域物体分界面的离散曲面基，实现了点集中曲面关键点计算、曲面构件提取与划分等方法；(3)提出并实现了基于离散曲面基的三维曲面轮廓表达，解决了离散曲面在组合过程中尖锐特征保留、连续性保留等问题；(4)以“多材质聚类-全局特征提取-离散曲面基构造-有向骨架树构建”为方法主线，建立了一套多域物体分界曲面提取与表达的方法体系；(5)提出了一种基于切片的点云修补方法和基于切片和线面拟合的人造物三维场景恢复方法。该课题已发表论文10篇、申请专利4项、登记软件著作权3项，培养博士研究生3名、硕士研究生4名。已全面完成了计划任务书中的各项指标。