【作者】 王坚；

【导师】 周来水；

【作者基本信息】 南京航空航天大学 ， 航空宇航制造工程， 2007， 博士

【摘要】 现实世界中存在着大量的碎片,碎片中有的极具价值,人们希望将它们拼合复原。采用计算机辅助技术进行碎片的拼合仿真能够大大提高碎片拼合复原的效率。本文对非规则碎片的拼合技术进行了深入研究,包括碎片数字化、全局和局部特征提取、碎片分类、碎片匹配和碎片拼接等关键技术,这些技术在计算机视觉、反求工程、临床医学等领域中也有重要应用。本文主要研究内容和成果如下:研究了三维碎片的数字化及碎片分类技术。使用光学扫描仪获取碎片点云数据,在数据测量和数据预处理的基础上,提出了基于STL文件建立碎片三角网格模型的算法。在基于碎片厚度分类的基础上,重点研究了旋转类碎片的分类算法,算法基于碎片的特征点和特征量,按照定义的分类规则将种子碎片归类,并使用母线匹配算法完成其它碎片的归类,形成碎片数据库作为后续拼合的基础。针对碎片拼合的共性基础——轮廓曲线匹配问题,提出了根据碎片轮廓曲线的形状特点分别匹配的策略,即基于特征点的匹配和基于轮廓分段的匹配。基于特征点的匹配算法首先通过样条拟合计算曲率,提取特征点,进行单个特征点匹配,然后构造相似矩阵,通过搜索相似矩阵得到满足相容性约束的多个特征点的匹配,提高匹配的可靠性。基于轮廓分段的匹配算法适用于具有角点而其它特征点较少的轮廓曲线,该算法在轮廓分段和粗匹配的基础上使用动态规划技术进行精匹配,在保证匹配效果的同时可提高匹配效率。针对薄壁类碎片(包括平面碎片和空间旋转类碎片),提出了基于轮廓曲线匹配与三元组融合的碎片整体拼合算法。该算法首先提取碎片边界轮廓曲线,使用轮廓曲线匹配算法进行碎片的两两匹配,使用基于三元组的整体拼合算法进行碎片的整体拼合。该算法的特点在于:在碎片两两匹配过程中自动判别轮廓曲线的走向,在整体拼合过程中综合运用整体配准、重叠检测和干涉检查剔除误匹配,以提高匹配可靠性。针对三维非薄壁类碎片,提出了基于曲面匹配和子图融合的多碎片整体拼合算法。该算法基于特征线进行区域分割提取拼合断面,基于snake模型对断面区域边界进行优化;基于最大权团的曲面匹配算法与遗传算法相结合的方法进行碎片初始匹配;基于子图融合法进行多碎片的整体拼合,在拼合过程中通过碰撞检测、位置一致性检测和跨界连续性检测排除误匹配,提高匹配可靠性。针对三维视图碎片,提出了一种不依赖于人工标记点的多视图整体拼合算法。该算法以视图碎片中高曲率点和平均采样点的自旋图作为局部形状标签进行初匹配,初匹配点对对集由最大权团搜索算法获得;基于最小生成树确定匹配视图对,采用改进的ICP算法获得视图对的精确匹配点对集;采用改进的基于奇异值分解的同时配准算法,根据精确匹配点对集进行多视图的全局拼合与优化。更多还原

【Abstract】 Lots of irregular fragment examples have existed in real world, some of which has great value, so people hope to reassembly them. Introducing computer aided technique to the field of fragment assembly has greatly speeded the assembly working process. This thesis has studied deeply on irregular fragment assembly methods, which include key techniques such as fragment digitalization, global and local feature extraction, fragment classification, fragment matching, and fragment assembly, these techniques can be employed in some other fields such as computer vision, reverse engineering, clinic medicine. Its main contents and contributions are as follows:3D fragment digistization technique and classification technique have been researched. First of all, optical scanner is used to to acquire the fragments’3D point cloud data. After data preprocessing, triangle mesh models of the fragments are constructed from STL files. Based on the proposed fragment classification methodology by thickness, the classification techniques targeting on revolved fragments are furtherly researched. According to the priori classification rule, which is based on feature point and feature metric, seed fragments are classified firstly. All the remained non-seed fragments are classified with profile matching method. A fragment library then has been constructed and used as the basis of the succesive fragment assembly work.Aiming at contour curve, a contour cuve matching strategy is proposed, which adopted two different matching methods according to the curve shape characteristic, namely, feature-based matching algorithm and segmentation-based matching algorithm. The feature-based matching algorithm estimates curvature by spline fitting, extracts the feature points, and obtain single-feature-point match, then constructed the similarity matrix, and searched the similarity matrix to obtain the multiple-feature-point match satisfying compability constraint. The algorithm improved the matching reliability. The segmentation-based matching algorithm has been used to contour curve with distinct corner point and obscured feature point. After contour segmentation and coarse matching, a dynamic algorithm is used to improve matching quality. The algorithm speeded the matching process and assured the matching quality.Aiming at thin walled fragment, a reassembly algorithm based on contour curve matching and triplet merging technique is proposed. Thin walled fragments included planar fragments and space revolving fragment. Firstly, the algorithm extracted boundary curve, used curve matching algorithm to do piecewise matching. Fragment assembly is completed based on triplets. The algorithm is characteristic of improving the reliability by auto-decision of curve direction and removal of fasle maching via simultaneous registration, overlapping detection and interference detection.Aiming at 3d non-thin walled fragments, a multiple-piece reassembly algorithm based on surface matching and sub-graph merging technique is proposed. The algorithm extracted the fracture region using feature-line-based segmentation method and optimized the frature region’s boundary by snake model. A piecewise matching algorithm is proposed, which is the combination of maximum-clique surface matching and genetic algorithm. A multiple-piece assembly is proposed based on sub-graph merging technique. Bumping detection, global consistant detection and cross-boundary continuity dectection in the assembly process are used to romove the false matching.Aiming at 3d view pieces, a reference-point-independant multi-view reassembly algorithm is proposed. The algorithm uses the spin-image of high curvature point and uniform sampling point to do initial piecewise matching. Based on maximum weight clique, the initial matching point set is determined. Base on the minmum spanning tree, an enhanced ICP registration algorithm is proposed to do view pair’s registration optimization and obtain the accurate matching point set. According to the accurate matching point set, an advanced SVD-based simultaneous registration method is proposed to minimize all the view-pair accumulated registration error.更多还原