【作者】 钱锦锋；

【导师】 叶修梓；

【作者基本信息】 浙江大学 ， 计算机软件与理论， 2005， 硕士

【摘要】 20世纪90年代以来，激烈的市场竞争对产品研制开发的时间和产品的更新换代速度提出了越来越高的要求，运用新的设计制造技术来满足市场需求层出不穷，其中在产品设计制造领域中广泛采用了逆向工程(Reverse Engineering，RE)来缩短产品研制时间。RE指的是根据现有的产品模型，利用数字化测量设备获取实体数据，然后对这些数据进行拟合，构建一个完整的CAD模型，因此逆向工程可以分为三个过程—数据采集(data acquisition)，数据处理(data processing)，抽象建模(abstract & model)。本文主要研究后二个方面的内容—即如何高效的处理、显示实体数据(数据处理的内容)以及后续的抽象建模—如特征的提取、曲面的拟合。 随着测量设备的数字化，自动化，精度程度的不断提高，模型的测量数据呈快速增长趋势。目前一般的激光测量设备可以从产品表面轻易获取数十万甚至几百万的测量数据。如此巨量的测量数据不仅加大了系统的负荷，而且大大降低了后续处理的效率。因此在抽象建模之前，对点云数据进行预处理显得尤其重要。本文将详细论述了点云预处理，其中包括的主要内容有：点云的采样、平滑滤波，点云的分割融合，坐标变换，数据的派生和重组，数据的排序和矢量化。在总结点云预处理的各方面内容目前处理方法和研究情况的同时，本文亦提出对一些算法上的改进，尤其是点云的采样中，本文提出了一种新的在大规模采样过程中保留边界点的算法。 预处理完后的下一步是抽象建模。本文中点云的建模主要包括特征线和特征面的提取。特征线的提取简单的有二种方法，一种是点云数据通过构造三角化网格，根据三角网格的拓扑关系可以检测出大略表达实体的特征线，另一种直接由点云数据拟合、数据排序构造出轮廓特征线—本文将详细论述这种方法。特征曲面的提取与特征线类似，有多种方法：可以以特征线为基础生成曲面，如放样法(lofting)、Bound UV Curve Network方法，也可以直接通过点云数据直接拟合曲面。前者曲面受特征线的方向影响很大，特征线的方向代表着曲面矢量方向，如果特征线方向不正确，往往会得到不正确的曲面形式，甚至完全失去实体原来的形状特征。而后者则相反，可以生成很高精度的曲面。因此本文详细论述由点云拟合成二次曲面的方法。本文的研究内容是详细论述了逆向工程中点云处理的主要内容:包括对测量数据的处理和抽象建模一特征线提取、曲面的拟合。 本文的主要贡献在于在分析、总结目前点云预处理的各方面内容后，尤其是点云采样，边界特征提取，二次曲面拟合等方面提出了一些创新的想法和以后的发展展望。关键字:逆向工程;点云;点云处理;采样;边界特征;二次曲面;曲面拟合今更多还原

【Abstract】 Since the 90s of 20th century, due to the violent market competition, higher and higher requirements for the time of product research and development and the speed of product updating are emerging. Brand new technologies on design and manufacture to satisfy market demands come out. Reverse Engineering (RE) is one of those wide-used technology to short the product research and development time. The definition of RE is that given a three dimensional real model, constructing a integrated Computer Aided Designed model which can be modified by three dimensional modeling tool like Solidworks, from the discrete space coordinate data of the real model. The digital measure equipment can used to acquire the discrete data also called point cloud. Therefore Reverse Engineering can divide into three processes: data acquisition, data processing, and abstract & model. In this paper we mainly discuss the later two processes: how to efficiently process, display and render the model data (data processing content) and the extraction of feature curve and the fitting of feature surface (abstract & model).With the development of digital measure equipment, the improvement of automatic acquisition and precision, the quantity of measure data from the model grows rapidly. General laser measure equipment now can easily acquire ten thousands even hundred thousands data. The data is so large that it not only burdens the system, but also reduces the efficiency of the later process. So it is very important to pre-process the origin data before the abstract & model phase. The pre-processing phase includes point cloud sampling, smoothing and filtering, segmentation and merging, coordinate transform, deriving and rearrangement, data sorting etc. During the introduction of these algorithms and research development, innovative or improved algorithm is put forward on some points. Especially the point cloud sampling, a new algorithm of preservation of boundary points while the large-scale compression is proposed and implemented. This paper will discuss the pre-processing later in detail.The next phase of pre-processing is abstract & model. In this paper, the modeling of point cloud includes the extraction of feature curve especially the boundary curve and fitting surface from scattered data especially the quadric surface. There are two simple ways in the extraction of feature curve. One is delaunay the point cloud to a mesh, and then detection the feature curve according to the topology relation of the mesh. Another way is extracting the feature curve directly from the point cloud by fitting and sorting the feature points will be discussed later in detail. Similarly, the extraction of feature surface has many ways. Surface can loft or apply the Bound UV Curve Network from the feature curve. Another way is independent on the curve. It constructs the feature surface by fitting the point cloud. The first method aims to free surface construction, but depends on the curve’s direction and will lost the original model shape if the curve direction which representing thesurface direction is incorrect. On the other hand, the later can make surface with high precision. Quadric surface fitting usually adopts the second way and will be discussed later.This paper aims to the research of point cloud processing in reverse engineering, including the pre-processing of point cloud and abstract model like the extraction of feature curve and the fitting of surface.The main contribution of this paper is that, creative idea is proposed especially in point cloud sampling, the extraction of boundary feature curve and quadric surface fitting based on analyzing and summarizing the current algorithm of the pre-processing. At last, some suggestions about future research are proposed.更多还原