【作者】 吴世雄；

【导师】 陈子辰； 王文；

【作者基本信息】 浙江大学 ， 机械制造及自动化， 2005， 博士

【摘要】 逆向工程在创新产品设计中起到日益重要的作用,如何有效而精确地获取复杂实体的三维数据是目前急需解决的问题。单一传感器逐渐难以满足复杂测量要求,而多传感器测量成为发展趋势。 本文提出多传感器集成智能化测量理论。多传感器测量系统集成视觉传感器、激光测头、接触式测头和数控装备。该集成测量系统能够利用多传感器的优势,达到快速智能化测量复杂实体的目标。论文主要包括以下内容:1.研究CCD视觉传感器、接触式测头及激光非接触式测头的基本测量原理和测 量方法。对于CCD视觉测量,采用一个CCD传感器实现立体视觉曲面测量; 对于接触式测量,分析基于三角细分曲面测量方法,提出基于矩形细分的未 知自由曲面自适应测量规划;对于激光非接触式测量,提出基于采样策略的 实体边界测量,以及基于变曲率的未知自由曲面自适应测量规划。2.研究了去毛刺、滤波、数据压缩等数据预处理方法,提出新的区域增长算法, 构建散乱点云的优化三角面片模型。首先提出简单的去毛刺处理方法及光顺 处理准则,并进行中值滤波处理仿真研究。为了避免数据量过大的弊病,提 出基于剖分小立方体的数据压缩方法。为了恢复散乱数据点之间的拓扑关系, 提出新的区域增长方法构建三角面片模型。在三角面片集合的区域增长过程 中,提出“最小边角积”法则搜索合适邻接点以形成新三角面片。3.提出散乱点云的特征智能识别理论,主要包括三步:恢复散乱点云的微分几 何属性、点云分割及特征识别,其中恢复点云微分几何属性是边界分割和特 征识别的基础。提出改进的Taubin方法,恢复散乱噪声数据的主曲率和 Darboux框架。利用边界点的微分几何特征,提出散乱点云自动分割方法, 有效提取噪声点云的D~0、D~1和D~2边界带。对分割后的点云区域建立曲率直 方图,达到快速有效确定点云曲面特征的目的。4.建立点云特征指导下的多传感器智能测量方法。点云特征分为曲面特征和边 界特征。利用点云特征规划测量的方法为:对于二次曲面点云进行采样处理, 对于自由曲面点云进行切片处理、截面线步长计算,最后进行Zigzag路径规 划以得到优化的测量路径。实体边界对于测量具有重要作用,提出边界特征 指导下的曲面及孔洞测量方法。利用规划好的测量路径,可指导高精度测头 快速智能化测量。5.研究了多传感器测量信息融合技术。首先整合多传感器多视觉测量数据,整 合后的数据进行曲面拟合及精度评测,达到精度要求的数据用来恢复曲面特 征。对二次曲面采用类型指定的最小二乘法拟合曲面,对自由曲面进行非均更多还原

【Abstract】 Reverse engineering plays an increasing important role in creative product design. At present, how to acquire 3D coordinate points of complicated objects efficiently and accurately is a critical issue to be solved. Single sensor cannot satisfy complex measurement requirements increasingly, and the multiple-sensor integrated system becomes a new development direction.This dissertation proposes an intelligent measurement theory of multiple-sensor integrated system. The coordinate-measuring system integrates vision sensor, laser scanner, contact probe and CNC machine. The integrated-system can take advantage of multiple-sensor to measure the complex objects in a fast and intelligent way. The main contents include the following aspects.1. Basic principles and measurement methods of CCD vision-sensor, laser scanner and contact probe are researched. A single CCD sensor is utilized to realize stereo-vision surface measurement. For the contact measurement, the triangle-division measurement way is analyzed first, and then a new method with the rectangle-division technique is proposed to measure unknown free-form surface adaptively. For the laser scanner, an edge measurement method based on sample points is put forward to measure object edges effectively, and a new measurement-planning method based on change-curvature is brought forward to measure unknown free-form surface adaptively.2. Several data-preprocess methods, burr elimination, filter, and compression, are studied. And a new area-growth method is presented to construct an optimized triangulated-surface model for unorganized data. A simple way of eliminating burr and a data-smooth principle is presented, and median-filter method is studied though emulation experiments. A compressing technology based on division-cube is presented to reduce the data number. To construct the topology relation of adjacent points, a new region-growing method is proposed to generate a triangulated-surface model from massive unorganized points. In the region-growing process of triangulated surfaces, a minimum-edge-angle-product algorithm is presented to select an appropriate point to form a new triangle.3. A theory of intelligent feature-detection for unorganized point-cloud is provided. It includes three steps: recovery of differential geometry property for unorganized points, data segmentation and feature detection. The first step is the base of the other two steps. The algorithm from Taubin is adjusted to estimate the principal curvatures and the local Darboux frame of unorganized noisy data. Using the differential geometry property of edge points, an automatic segmentation algorithm is presented to extract the D~0, D~1 and D2 edge strips from noisy data. For each segmented point-cloud patch, a principal curvature histogram is constructed to identify its surface feature effectively.4. Based on the detected feature of point-cloud patch, an intelligent measurement method with multiple sensors is proposed. The feature of point-cloud patch includes surface feature and edge feature. To plan surface measurement path from the surface feature, first sample a few points from the quadratic surface, or sliver the point-cloud and compute measuring steps foreach slivered curve. Then compute Zigzag path to acquire optimal measurement path. The edge of object is very important for measurement. An adaptive surface and hole measurement method directed by edge feature is provided. Based on the planned measurement path, high-precision sensors, touch probe or laser scanner, can be guided to measure the object in a fast and intelligent mode.5. The information fusion technique of multiple-sensor measurement is studied. The measured points of multiple views are registered firstly. Then surface reconstruction and surface accuracy assessment are performed. Surface patches that satisfy the accuracy requirements are used to recover surface feature. A type-specified least-square method is developed to reconstruct quadratic surface. To reconstruct free-form surface, B-spline curves fitting is performed and NURBS representation is generated. The measured points and surface representation are transformed into standard IGES format, which can be imported into commercial CAD/CAM software. The massive unorganized points can be used to generate optimal STL file directly through data compression based on triangulated surfaces.6. The intelligent measurement system that integrates multiple sensors is researched. The hardware frame of the multiple-sensor integrated measurement system includes motion controller system, accurate servo system, accurate mechanical system and multiple-sensor. The main functions of measurement software are introduced, in which intelligent measurement module is the key of the whole system. The intelligent measurement system that integrates multiple sensors are testified and verified by practical experiments.更多还原