基于计算机视觉的镜面体表面三维测量技术研究
文献类型：学位论文
| 作者 | 付生鹏
|
| 学位类别 | 博士 |
| 答辩日期 | 2015-05-27 |
| 授予单位 | 中国科学院沈阳自动化研究所 |
| 授予地点 | 中国科学院沈阳自动化研究所 |
| 导师 | 刘伟军 ; 赵吉宾 |
| 关键词 | 计算机视觉 三维外形测量 镜面体 |
| 其他题名 | Specular Surface Measurement Based on Computer Vision Technique |
| 学位专业 | 机械电子工程 |
| 中文摘要 | 目前，物体的三维外形测量技术在工业设计、产品质量控制、逆向工程、文物保护、医疗卫生等领域都有十分广泛的应用。物体三维外形测量技术的测量精度直接关系到所获取的物体三维模型的精度。在实际的测量需求中，有一类表面光学性质特殊的物体被称为镜面物体，这类物体表面光滑，对光线的反射呈现镜面反射的状态。由于其表面的特殊性，在使用接触式测量时，容易划伤物体表面，并且测量效率低下。计算机视觉测量技术作为一种非接触式测量，具有测量速度快、方便灵活的优点，在测量领域应用越来越广泛。但是目前的成熟的视觉测量技术通常只能完成表面呈漫反射状态的物体的测量。虽然广大学者对基于视觉的镜面体测量进行了深入研究，但目前的测量方法均存在着很多的问题，不能准确的对镜面物体完成测量。因此，本文在自然科学基金项目“基于数字相移和双目视觉的非朗伯体表面测量方法(51005229)”以及横向课题“航空透明件加工与自动化研抛技术研究”的支持下，展开了基于计算机视觉的镜面体表面三维外形测量技术的研究。 本文对现有的基于计算机视觉的镜面体测量技术进行了研究，客观分析了现有方法的优缺点，根据镜面体表面对光线特殊的反射特性，设计了一套基于双目视觉的镜面体三维外形测量系统。该系统由一台计算机、一台LCD显示器，两部相机以及一个电动转台组成。由计算机产生编码条纹，控制LCD显示器向镜面物体表面投射条纹图案，经过镜面物体的反射后，由相机获取图像。最后根据条纹图案经物体反射后产生的形变信息，对测量区域进行了三维重建。由于在转台上粘贴了人工标志点，使用双目测量的方法，计算出标志点在相机坐标系下的坐标，在建立起所有标志点的一一对应关系后，可以获得转台在不同角度时相互之间的旋转平移关系。根据计算出来的旋转平移关系，可以将转台在不同位置测量获得的物体表面三维点云数据统一到一个坐标系下，从而实现了对物体表面的完整测量。 在测量过程中，由于需要LCD显示器投射条纹图案，由相机获得镜面物体反射的条纹图像，对条纹变形进行分析，从而恢复物体表面的三维外形，因此，需要对显示器的像素进行编码。在分析了现有的结构光编码方法的基础上，采用了一种格雷码和相移法相结合的编码方法，该编码首先将显示器划分为几个小型区域，使用格雷码对划分的每一个小区域进行唯一性编码，然后在每一个小区域内再使用相移法进行编码。对两种编码分别进行解码，在使用相移法计算得到截断相位值之后，将该相位值加上该区域的格雷码值，就可以得到连续的绝对相位值。这种编码方法既发挥了格雷码可以实现表面不连续的物体的测量和相移法对噪声不敏感测量精度高的优点，又减少了需要投射的条纹图案的数量。 在使用该系统对镜面物体进行三维测量前需要首先进行系统的参数标定。由于使用了两台相机，因此系统的参数标定包括相机的内部参数标定以及相机的外部参数标定，其中外部参数标定还包括相机-相机之间外部参数的标定以及相机-显示器之间的外部参数标定。由于显示器不在相机的视野范围内，因此需要使用一个形状规则的镜面对显示器进行反射，使得相机能够获得显示器的图像，然后根据镜面成像的约束条件完成相机和显示器之间位置关系的计算。通常使用的镜面有普通平面镜和球形镜面，但是两种镜面均存在缺点。在分析现有标定方法的基础上，提出了一种基于环形镜面的相机-显示器外参标定方法。该方法只需拍摄一次镜面反射的图像，即可完成外参标定，最后通过仿真和实际实验验证了该方法方便快捷，具有较高的标定精度。 现有的对镜面体进行准确测量的方法一般都是单点测量，这种测量方式容易引起较大误差。本文分析了表面光滑且连续的镜面体表面的几何特性，指出了其表面存在的光滑性约束和理想成像约束，从这些约束性质可以看出镜面体表面一点的法向与周围各点之间存在的约束关系，并不是孤立的。因此，从整体测量的思想出发，使用一个光滑曲面代替镜面物体，通过不断调整光滑曲面的形状，模拟实际物体的成像过程。当模拟成像与实际成像之间的误差最小时，我们就认为此时的光滑曲面与待测镜面物体的三维外形相同。由于实际的LCD显示器发出的光线的方向无法直接获得，因此实际的计算过程中采用光线逆跟踪的方法，并且使用双三次B样条曲面作为光滑曲面，以方便调整曲面的形状。通过测量实际实验，验证了该测量方法具有较高的测量精度。 本文对镜面物体的测量系统设计、系统的参数标定、投影装置的编码和图像解码、镜面物体的三维重建方法进行了深入的研究。这些工作为实现镜面物体的高精度测量奠定了基础，同时分析了测量过程中存在的问题，指出了镜面物体三维外形测量下一步的主要研究内容和方向。 |
| 索取号 | TP391.41/F97/2015 |
| 英文摘要 | At present, the 3D shape measurement technique is widely applied in the areas of industrial design, product quality control, reverse engineering, protection of cultural relics, health care and so on. The measuring accuracy of the 3D models of the objects depends on the measurement technique. In the actual measurement requirement, there is a certain type of objects with special optical characters called specular surface. The surfaces of this kind of objects are smooth and can reflect all of the incident lights like a mirror. If the contact measurement method is used to measure this kind of objects, the surface may be destroyed and it will take a lot of time to complete the measurement. The computer vision based measurement technique which is fast and noncontact is applied widely in nowadays. But the traditional vision based measurement instruments are fit to measure the objects of diffuse surface. To measure the specular surface precisely, a lot of researchers have developed kinds of measurement methods. But there are many problems to be solved. This research focuses on the study of the specular surface 3D measurement base on computer vision technique with the support of the NSFC “Phase shifting and binocular vision based non-lambert surface 3D measurement” and the project “Research on processing and automatic polishing technique of aviation transparent parts”. The existing specular surface measurement methods based on computer vision technique are studied in this paper. A measurement system consisting of a computer, an LCD display, two cameras and a turntable is designed based on the analysis of the principle and disadvantages of the existing measurement methods. To measure a specular surface, the LCD display is used to project encoded fringes, controlled by the computer, so that the camera can capture the images of the fringes reflected by the specular object. The 3D shape of the object can be obtained after the analysis of the distortion of the fringes. Some circle points are attacked to the top surface of the turntable. The two cameras capture the images of the points and reconstruct their 3D coordinates. A complete shape of the specular surface can be formed after all of the 3D data are transformed into a unit coordinate system. An LCD display is used to project encoded stripes. A lot of stripes encoding methods are studied in recent years. On the analysis of the existing structured light encoding methods, a method combined gray code and phase shifting method is adopted. The whole coding area will firstly be divided into several small areas. We use the gray code to encode every small area with a uniqueness code, and the phase shifting method will be applied in every small area then. After decoding the two stripes respectively, the absolute phase value is calculated by plus the truncated phase value and the gray code value. This encoding method not only takes the advantage of the gray code that it can measure a discontinuous surface, but also takes the advantage of the phase shifting method that it is not sensitive to the noise and can achieve high measuring accuracy. This method can also reduce the number of stripes to be projected. The measurement system should be calibrated at first. The calibrations include the intrinsic calibration of the two cameras and the extrinsic calibration. The extrinsic calibration contains the extrinsic calibration between two cameras and the calibration between the camera and the LCD display. Since the display is not in the field of view of the camera, so the calibration between the display and the camera must be carried out using a planar or sphere mirror. But there are some problems to use these two kinds of shapes of mirror. For example, there will be a degenerated case by using a planar mirror. A sphere mirror is difficult to machining and the calibration accuracy depends on its radius. So an annular mirror based extrinsic calibration method is proposed. The annular mirror is just needed to place at one position and the parameters are obtained. The simulation and real experiments results show that the calibration method is effective and obtains high accuracy. Most of the existing methods for specular surface measurement depend on the point by point measurement. This paper analyzes the surface smooth and continuous geometry characteristics on the specular surface, and points out its surface smoothness and ideal imaging constraints can be used to reconstruct the normal and 3D depth of the object. Therefore, based on the idea of overall measurementa simulation method of the actual object imaging process is used to reconstruct the object through continuous to adjust the shape of a smooth surface. When the errors between the simulation and actual images reach the minimum, we think the smooth surface and three-dimensional shape of the mirror the object are the same. Because the actual the direction of the light from the LCD display can't directly accessed, so in the process of the actual calculation, the method of light inverse tracking method is adopted. A bi-cubic B spline surface is used as a smooth surface and high accuracy is obtained after the experiment. In this paper, the design of the measurement system, the parameters’ calibration, the encoding and decoding method of the LCD display and the reconstruction method of the specular surface are studied. This work laid a solid foundation for the realization of high precision measurement of specular objects. The problems existing in the measurement process are analyzed at the same time. To realize the measurement of mirror objects of arbitrary shape is the main research content and direction of the next step. |
| 语种 | 中文 |
| 产权排序 | 1 |
| 页码 | 121页 |
| 源URL | [http://ir.sia.ac.cn/handle/173321/16764]  |
| 专题 | 沈阳自动化研究所_装备制造技术研究室 |
| 推荐引用方式 GB/T 7714 | 付生鹏. 基于计算机视觉的镜面体表面三维测量技术研究[D]. 中国科学院沈阳自动化研究所. 中国科学院沈阳自动化研究所. 2015. |
入库方式： OAI收割
来源：沈阳自动化研究所
浏览0
下载0
收藏0
其他版本
除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。