【作者】 孟娜；

【导师】 周以齐；

【作者基本信息】 山东大学 ， 机械制造及其自动化， 2009， 博士

【摘要】 随着计算机技术和机械制造技术的日益发展,逆向工程已广泛应用于产品再创新和设计中。作为产品零部件外形几何逆向工程的一个研究热点,基于激光扫描点云的数据处理技术作为逆向工程几何建模的重要技术,它以激光扫描点云作为逆向数据预处理和建模的基本元素,目前该技术在国内外得到蓬勃发展。该项技术以获取的点云数据为处理对象,不用构建三角网格,在处理超大规模点云中,对点云数据进行预处理、特征提取和模型重构方面,显示出其独特的优势,现在正成为逆向工程研究的一个热点。本文针对该领域中的若干关键性问题,结合山东省自然科学基金项目“基于多尺度特征的复杂曲面重构技术研究”(项目编号:Y2006F12)进行了深入地研究。(1)为了满足逆向工程后续产品开发和重构的精度要求,本文完成了激光扫描点云数据的噪声数学描述与分类,制定了把各种噪声降到最低甚至有效去除的数据预处理流程。根据激光扫描点云数据的特点,给出了噪声点的产生机理及其数学描述,根据建立的数学模型,把噪声点进行分类:由系统测量误差α(x_i,y_i,z_i),和系统随机误差β(x_i,y_i,z_i)引起的噪声点,以及由随机性分量g~S(x_i,y_i,z_i)引起的噪声点。分别根据其特点制定切实可行的去噪方案有针对性的去除。为此,制定了一套数据预处理去除噪声的流程,包括明显噪声点去除,噪声点滤波平滑处理,点云数据的光顺处理等。工作结果表明,在求得的切片点云(从模型的三维离散点云数据到获取的二维截面轮廓的点云数据)基础之上,所采取的去除噪声预处理流程能够把由系统测量误差、系统随机误差以及随机性分量所引起的噪声点减小甚至部分消除,满足逆向工程后续产品开发和重构的精度要求。(2)研究了激光扫描点云数据的有关预处理算法,提出了优化修正量光顺算法。本文在激光扫描点云数据密集、数据量大、不便于存贮的背景下,在保证精简后数据精度的前提下,提出一种用偏差参数和角度允差值来进行点云数据压缩的算法,该算法用于处理的激光扫描点云数据是经过大噪声点去除、滤波和优化修正量光顺处理之后的海量点云数据。该压缩算法简单直观,能够根据公差值d_T和角度最大允差值的大小来压缩数据,这一点能最大限度地满足机械产品外形和精度要求。能够最大限度地保留原有点云数据的外形,提高压缩后数据点的精度,对海量点云数据的压缩具有实际应用价值。针对所采集的激光扫描点云密集和数据量大的特点,重点研究了截面线点云数据的光顺处理,这方面比较著名的文献包括,Eck M.,Jaspert R.和G.H.Liu,Y.S.Wang and Y.F.Zhang所提出的光顺算法,可是,G.H.Liu和Y.S.Wang等人所提出的光顺算法,其修正量是递进的,利用寻优函数的约束来确定修正量,从而使坏点得到光顺,其修正量的阈值没有限制;修正方向的指向是按照能量函数方程符号的变化作出决定,其编程实现相对复杂。由此,本文提出了一种优化修正量光顺算法,在分块进行粗、精光顺处理采样数据过程中,分别由曲率及其一阶差分符号的变化来辨识坏点。坏点的修正方向直接按照能量函数方程确定出由型值点指向三角形形心的正或负的G向;修正量由赋初值开始,然后按照能量函数方程,递进搜索,满足能量代数式最小值后搜索停止。本文所提出的优化修正量光顺算法,主要用于光顺激光扫描散乱点云数据,该算法能够满足曲线曲面重构的光顺性要求,可以有效保留曲线的原有几何外形。最后通过在二维散乱点云上的实例仿真,验证了所提算法的适用性和有效性。(3)提出了用离散曲率算法进行截面线特征提取。特征提取是逆向工程的重要步骤,其中截面线特征点的弱化是需要解决的关键问题。本文重点研究了二维截面线特征点的提取。检索到有关特征点直接提取的文献有,自适应k-曲率(AKC)函数算法,在断点提取中,AKC函数是用于提取拐角和光滑连接之间的特征点;映射高度函数(PHF)算法,PHF函数用于从圆弧中区分出直线段的特征点提取;由Liu和Ma提出的相对转角绘图(RSTM)算法,用于辨识轮廓线的特征点提取问题。AKC函数算法和PHF算法只能提取某种特征点,其广泛应用受到一定限制。在研究了特征点直接提取上述文献相关算法的基础上,提出用离散曲率法提取特征点。所提出算法的主要内容包括:用包含了高斯核函数曲线的曲率表达式建立相关数学模型,选用了合适的离散尺度因子。根据离散曲率曲线的局部极值点,确定出截面线特征点集,并进行特征点的融合。所提出的算法用于准确地获取激光扫描点云的原始设计意图,能最大限度地与原有形状特征元保持一致。顺利地完成逆向建模过程的关键一步。在实例应用中,把RSTM算法和所提出的离散曲率算法在实例应用上做了输出比较,结果是,本文提出的离散曲率算法特征点提取问题,能够提取弱化的特征点,不容易出现特征点的漏检问题,是一种适用和有效的算法。更多还原

【Abstract】 With the development of computer and mechanical manufacturing technology, reverse engineering has been widely used in product re-innovation and design. As a research hot topic of geometrical components and an important geometric modeling technique in reverse engineering, data processing technology based on laser scanning point cloud, which is regarded scattered point cloud data as the the basic element during the process of data pre-processing and modeling, and which is so important that it is full of development at home and abroad at present time. The data processing technology, regarded the obtained point cloud data as processing objects and without building a triangular mesh, now shows its unique advantages and is becoming a hot research spot, during the process of dealing with very large scale point cloud, point cloud data preprocessing, feature extraction and model reconstruction. In this paper, some key issues in the field of reverse engineering have been developed deeply, which is helped by the National Natural Science Foundation of Shandong Province " Complicated Surface Reconstruction Technology Research Based on Multi-scale Features " (Item Number: Y2006F12).(1) In order to meet the precision of following product development and reconstruction in reverse-engineering, mathematical description and classification about noise have been completely finished, and a set of data pre-processing process has been set up, which can minimize or effectively eliminate the noise.According to the characteristics of laser scanning point cloud data, some mathematical description and generated mechanism of noise points are descriped. Based on the established mathematical model, the noise points are classified into two categories, one is that caused by the system measurement errorα(x_i,y_i,z_i)and the system random errorβ(x_i,y_i,z_i), the other is that caused by the random component g~s(x_i,y_i,z_i), the removal of which is carried up according to their characteristics relied on some feasible de-noised methods. So, a set of de-noised process of data pre-processing is developed, including the removal of obviously noise points, smoothing filter process of noise points, and smoothing process of point cloud data, etc.. The run results show that the de-noised pre-processing process can minimize noise points, and can meet the precision of following product development and reconstruction in reverse-engineering based on the sliced point cloud data (from three-dimensional discrete point cloud data of the model to the obtained point cloud data of two-dimensional cross-section contour), which caused by system measurement error, system random error and system random component.(2) The preprocessing algorithms about massive laser scanning point cloud data have been studied, and an optimized amount of smoothing algorithm has been proposed.In this paper, an bias parameter algorithm and an allowable difference of angle is presented to compress point cloud data, under the condition of intensive and massive laser scanning point cloud data which is not easy to store, and data accuracy assured after data compression, which is used to process laser scanning point cloud data through the big noise points removal, filtering and optimal amount smoothing process. The above mentioned compression algorithm is simple and intuitive, which can compress the data based on the size of the tolerance values about bias parameter and bias angle, and can meet the required appearance and precision of mechanical products greatly. The algorithm can preserve the original shape of point cloud data, and can improve the accuracy of the compressed data points, which has practical application value on the compression of massive point cloud data.As the collected laser scanning point cloud data is intensive and large, some smoothing process algorithm about laser scanning point data of a sectional curve have been mainly researched. The well-known literatures in this field, Eck M., Jaspert R., and G.H. Liu, Y.S. Wang and Y.F.Zhang, in which an smoothing algorithm is proposed. However, the smoothing algorithms are proposed by G.H.Liu and Y.S.Wang etc., the revised amount is progressive, and the constrained optimizing function is used to determine the amendment amount, which has no limit to the threshold value. And the revised direction to point is determined in accordance with changes of energy function equation symbols, in which the its program is relatively complex. Thus, an optimized amount of smoothing algorithm is presented in this paper, which identified bad points in accordance with the sign change of their curvatures and corresponding first-order differences during the process of coarse smoothing and fine smoothing. The corrected direction of bad points is pointed to the positive or negative G of the triangle centroid from the sampled data in accordance with the energy function equation. The revised amount has been an incremental search, beginning with an initial value and then following the energy function equation, until the minimum value of energy algebraic expression is meet. The proposed optimized amount smoothing algorithm in this paper, mainly used for smoothing scattered laser scanning point cloud data, can meet the smoothness requirements of curve and surface reconstruction. The proposed algorithm is particularly effective in terms of shape preservation. Case studies are presented that illustrate the efficacy of the proposed algorithm.(3) A discrete curvature algorithm is proposed to extract feature points of a sectional curve.Feature extraction is an important process in reverse engineering, in which weak feature points of a sectional curve is the key issues to be dealed with. The feature points extraction of two-dimensional cross-sectional curve is focused on in this paper. The retrieved literatures about feature points extracted directly include the adaptive k-curvature (AKC) function algorithm, which is used to extract the feature points between corner and smooth connection, the mapping height function (PHF) algorithm, which is used to distinguish feature points from arc and line segments, and the relative angle mapping (RSTM) algorithm proposed by Liu and Ma, which is used to identify the feature points of contours. The AKC function algorithm and PHF algorithm can only extract some certain feature points, which has certain restrictions used in the wide field.Based on directly extracting feature points of the above-mentioned documents, the discrete curvature method is proposed to extract feature points. The main contents of the proposed algorithm include, the curvature expression comprised of Gaussian kernel function curve is used to establish the relevant mathematical models, and a suitable discrete scale factor is chosen. According to the local extreme points of discrete curve, a set of feature points is determined, and the fuse of feature points is carried out subsequently. The proposed algorithm is used to accurately obtain the original design intent of the laser scanning point cloud, which can greatly keep with the original shape feature cell consistently. Then a key step in reverse modeling process is successfully completed. During the course of example application, an output comparison between the RSTM algorithm and the proposed discrete curvature algorithm has been carried out, which the the result is that the proposed method in this paper can extract the weak feature points, and cannot prone to undetect feature points. Then the output results show the proposed algorithm is practical and effective.更多还原