基于断层轮廓成像的三维模型重构理论和试验研究

发布时间：2019-03-31 16:40

【摘要】：随着制造业的持续快速发展,传统的正向工程已经难以满足人们的需求,反求工程越来越广泛的被应用到新产品开发设计中。反求工程不仅可以大幅度缩短新产品的开发设计周期,也可以大大降低研发的成本。在反求工程中,快速、准确地获取目标物体的三维几何数据是极为关键的一步,但是对于具有复杂内腔轮廓的零部件,数据测量方式受到了相应的限制,从测量精度和成本两个方面考虑,本文提出了一种基于断层轮廓成像三维模型重构的方法进行数据采集,是将CCD相机通过连接件与数控加工中心集成为一体,安装高分辨率的镜头与LED环形光源,并配有可拆卸的遮光桶,将用包埋体包裹后固定于机床工作台上的被测工件,使用数控加工中心进行高精度逐层切削成像,得到断层轮廓图像。之后对断层图像进行图像预处理、边缘检测、亚像素细分等技术进行处理,从而得到被测件的三维点云数据,再对三维点云数据进行点云精简、分割、三角网格化处理,最后通过曲线曲面拟合得到被测物体的三维重构模型。本文研究内容为:(1)CCD获取断层轮廓图像的试验研究通过搭建试验平台,对被测件进行预处理之后,根据试验设备的工作参数来选择合适的拍摄条件进行数据采集试验,通过对比和分析试验结果,确定最佳的拍摄条件以获取高质量的断层图像。在确定了拍摄条件的基础上,进行完整的三维模型重构试验的图像采集。本文共采集断层轮廓图像249层,共2490张。(2)图像的预处理以及边缘检测技术首先根据采集的原始数据,研究了图像的预处理技术,即平滑去噪、灰度平均法以及中值滤波;然后,通过研究已有的边缘检测算法canny算法和log算法,在此基础上提出一种改进的canny算法,一定程度上提高了边缘检测精度;最后通过亚像素细分算法进一步提高边缘检测精度,相比较像素级的边缘检测精度,亚像素细分算法精度提高了50%。(3)点云数据的处理研究了点云数据的处理,包括点云数据的精简、点云数据的分割和点云数据的三角网格化等。首先研究了点云数据精简的方法,如均匀采样法、包围盒法以及基于网格的方法,本文采用了一种基于曲率的点云精简算法,并通过实例验证了此算法的可行性;最后研究了网格三角剖分算法,提出了一种新的点云数据三角网格化算法。(4)曲面拟合和三维重构对曲面拟合进行了研究,包括曲面拟合理论基础,nurbs曲面拟合以及曲面的拼接。使用imageware、ug软件用基于特征曲线的重构、轮廓线叠加拟合重构和nurbs曲面拟合重构三种不同方法对经处理的三维点云进行曲面拟合,得到了被测件的三维重构模型,通过对重构误差的分析得到nurbs曲面拟合重构的重构精度最高。最后从数据采集、边缘检测精度和曲面重构精度三方面提出了改进重构精度的集中可行性方案。
[Abstract]:With the continuous and rapid development of manufacturing industry, the traditional forward engineering has been difficult to meet the needs of people. Reverse engineering is more and more widely used in the development and design of new products. Reverse engineering can not only greatly shorten the development and design cycle of new products, but also greatly reduce the cost of R & D. In reverse engineering, obtaining the 3D geometric data of the target object quickly and accurately is a very important step, but for the parts with complex inner cavity contour, the measurement method of data is limited accordingly. Considering the accuracy and cost of measurement, this paper presents a method of 3D model reconstruction based on fault contour imaging for data acquisition, which integrates CCD camera with NC machining center by connecting parts. Installed with high resolution lens and LED ring light source, and equipped with removable shading barrel, will be wrapped in embedded body and fixed on the machine tool workbench measured workpiece, using numerical control machining center for high-precision layer-by-layer cutting imaging. The contour image of the fault is obtained. Then the image pre-processing, edge detection and sub-pixel subdivision are carried out to obtain the 3D point cloud data, and then the 3D point cloud data is simplified, segmented and triangulated. Finally, the three-dimensional reconstruction model of the measured object is obtained by curve and surface fitting. The research contents of this paper are as follows: (1) the experimental research of CCD to obtain the image of fault contour by building a test platform, after pre-processing the tested parts, according to the working parameters of the test equipment to select the appropriate shooting conditions for the data acquisition test, Through the comparison and analysis of the experimental results, the best shooting conditions are determined to obtain high-quality tomography images. On the basis of determining the shooting conditions, a complete 3-D model reconstruction experiment of image acquisition is carried out. In this paper, a total of 2490 slices of slice contour images are collected. (2) pre-processing and edge detection techniques of images are firstly studied according to the original data collected, namely smoothing denoising, gray-level averaging and median filtering. Then by studying the existing edge detection algorithms canny algorithm and log algorithm an improved canny algorithm is proposed which improves the edge detection accuracy to a certain extent. Finally, the sub-pixel subdivision algorithm is used to further improve the edge detection accuracy. Compared with the pixel-level edge detection accuracy, the sub-pixel subdivision algorithm improves the accuracy by 50%. (3) the point cloud data processing is studied. Including point cloud data simplification, point cloud data segmentation and point cloud data triangulation and so on. Firstly, point cloud data reduction methods, such as uniform sampling method, besieged box method and grid-based method, are studied. A curvature-based point cloud reduction algorithm is adopted in this paper, and the feasibility of this algorithm is verified by an example. Finally, the mesh triangulation algorithm is studied and a new triangulation algorithm for point cloud data is proposed. (4) the surface fitting is studied by surface fitting and 3D reconstruction, including the theoretical basis of surface fitting. Nurbs surface fitting and surface stitching. By using imageware,ug software, three different methods of reconstruction based on characteristic curve, superimposed contour fitting and nurbs surface fitting are used to fit the surface of the processed 3D point cloud, and the three-dimensional reconstruction model of the tested part is obtained, which is based on the feature curve reconstruction, the contour superposition fitting reconstruction and the nurbs surface fitting reconstruction. The reconstruction accuracy of nurbs surface fitting reconstruction is the highest by analyzing the reconstruction error. Finally, from three aspects of data acquisition, edge detection accuracy and surface reconstruction accuracy, a centralized feasible scheme to improve the reconstruction accuracy is proposed.
【学位授予单位】：太原理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【相似文献】