【作者】 武剑洁；

【导师】 钟毅芳；

【作者基本信息】 华中科技大学 ， 机械设计及理论， 2004， 博士

【摘要】 基于点的点云处理技术是随着数据测量技术的进步而迅速发展起来的一门新兴技术。该项技术以点作为曲面绘制和造型的基本元素,在提高模型绘制与重建的速度、加强处理超大规模点云的能力和简化计算量等方面体现出独特的优势,目前已成为反求工程的一个研究热点。本文针对该项技术中的若干关键问题,结合国家十五重点科技攻关项目“产品设计 CAD”(项目编号:2001BA201A02)进行了深入研究。点集简化是点云处理中首要的预处理环节。为尽量避免损失被测物体的工程信息,提出了一种基于模糊聚类的简化算法。通过引入几何相似性隶属度来表征被测物体形状的自然变化,使简化点集倾向于聚集在曲面的陡峭区域,降低简化可能导致的形状损失;同时以强制约束相似性隶属度反映设计者的工程和设计要求,能有效抑制工程信息和设计信息的缺失。特征线提取是点云处理中另一项重要的预处理工作。为保证特征线提取的稳定性及精度要求,提出了一种基于数字图像薄化的多尺度特征线提取算法。利用局部熵和重复度描述采样点在不同尺度下属于某个特征的可能性大小,保证算法的稳定性;通过将提取的特征点云映射为数字图像和进行薄化处理,获取光滑的特征线,能在一定程度上处理密度分布不均的点云,并保证特征线的质量。曲面重建是点云处理的核心。为加快曲面重建的前处理速度,避免因采样点的少量丢失而导致重建表面出现缝隙,探讨了一种基于曲面单元分解的重建算法。借鉴推进波前法,基于子域环的构造和中心环的推衍,使所有点的法向指向被测物体的外侧,可缩短法矢检验的时间;在各采样点处建立相互交迭的曲面单元,以近似包围在点云内部的空间,可得到质量良好的重建表面。曲面编辑是点云处理不可缺少的研究内容。为获取灵活的曲面编辑能力,设计了一种基于超二次曲面的约束变形算法。建立基于超二次曲面的一般约束变形模型,使物体在多种类型的约束下按指定的曲线位移产生多样化的变形;采用局部熵加权的方式,使采样点能自动估计变形的程度,并在采样不足的区域插补适当数目的点,保证点云在变形前后的采样率基本一致。在上述理论研究成果的基础上,研制开发了基于点的三维散乱数据处理原型系统,并已作为一个模块嵌入三维CAD系统-Intesolid中。更多还原

【Abstract】 Advances in 3D scanning technologies have promoted the emergence and rapiddevelopment of point-based techniques. Point-based techniques, by which points are used assurface modeling and rendering primitives, has become an important research field of reverseengineering. The particular dominance of point-based techniques includes the efficiency atreconstructing and rendering very complex objects and environments, capability of dealingwith dense scattered point cloud, and simplicity of rendering algorithms. Based on theoverview of point-based techniques, several key issues including data reduction, feature lineextraction, surface modeling and editing are studied in this dissertation, which is sponsored bythe National Key Research Project of the 10th five-year-plan of China (Grant No.2001BA201A02). Data reduction is the first preprocessing step of point-based data treatment. To avoid lossof engineering information hidden in the measured object, a data reduction algorithm on thebasis of fuzzy clustering analysis is proposed. By introducing geometric similaritymembership, surface variation can be naturally represented, forcing samples to gather inregions where surface varies drastically. And imperative constraint similarity membership isintroduced to reflect engineering demand of designers, which is in favor of retainingengineering detail features. Feature line extraction from a point cloud is another necessary preprocessing step forsurface reconstruction. To satisfy the demand for stability and accuracy of feature lineextraction, a multi-scale feature line extraction algorithm based on digital image thinning ispresented. Local entropy and repeatability rate are introduced to classify points according tothe likelihood that they belong to some feature at different size of local window, whichachieves robust and stable feature point detection for noisy surfaces. By mapping theextracted feature point cloud into 2D digital images and thinning the images, smooth featurelines are recovered. Scanty data can be dealt with and the top-quality feature lines can berecovered. Surface reconstruction is the key problem of point-based data treatment. To save time onsurface reconstruction and avoid gaps of the reconstructed surface when parts of the data getlost during transmission, a new surface reconstruction algorithm by use of decomposition of II<WP=6>surface elements is discussed. Similar to the theory of Advancing Front Method, theconsistent tangent plane estimation of each sample is performed by constructing local loopsand advancing the center loop. The volume enclosed by the given point cloud is approximatedwith the set of overlapping surface elements built at each sample point. Surface editing is indispensable in point-based data treatment. To achieve a flexiblecapability for surface editing, a superquadric-based general constrained deformationalgorithm is designed. By building a superquadric-based general constrained deformationmodel, surface can be deformed according to user-specified curvilinear displacement underconstraints, which can consist of points, lines, surfaces and volumes. During surface editing,new samples are inserted to the original point cloud and located in proper positions by usingweighted local entropy method to preserve the overall sampling density. On the basis of the above theoretic achievement, a point-based data treating system from3D unorganized data points is developed and embedded in 3D CAD system - Intesolid.更多还原