高动态范围光亮表面的结构光三维形貌测量方法研究与实现

发布时间：2018-06-30 03:19

本文选题：三维测量 + 结构光　；参考：《广东工业大学》2017年博士论文

【摘要】：结构光三维测量技术目前在工业检测/质量控制、逆向工程(复杂自由曲面的数字化)、物体识别、文物保护、医学和虚拟现实等领域得到了广泛应用。通过该技术可以准确获得工件的几何形貌信息,为先进制造、自动装配、表面检测等提供有效的指导;另一方面,将工件的三维点云与CAD模型或已有三维数据进行比对,可检测出工件的形变,为质量控制、应力分析、碰撞测试等提供更加完整和更容易理解的可视化分析手段。然而,现有的结构光三维测量技术在应用中仍然存在一些问题,比如要求测量时环境光照限制在一定范围内,被测物体必须是漫反射表面,且表面反射率变化范围不大。而工业生产中有大量金属工件以铣削加工为主,经过加工处理后工件的表面会变得十分光亮,若直接对这种光亮表面进行测量,则镜面反射之后的光会太强,造成相机图像传感器饱和,丢失条纹图像高光区域的条纹信息,以致难以正常进行测量;另一方面,由于相机的动态范围有限,对于反射率较低的区域,则会造成条纹过暗,大幅降低测量精度。针对这些问题,本文系统地研究了结构光图案编码、相位误差补偿、系统标定和高动态范围条纹图像的获取等多个关键技术,提出一种自适应、高动态范围结构光三维测量新方法,包括线移法编码结构光图案、基于特征点映射的系统标定算法、自适应调节条纹图案的最佳投射灰度值和基于平滑样条拟合的相位误差补偿算法,有效地解决了光亮表面结构光三维测量中数据丢失的难题,为高动态范围光亮表面的三维测量,特别是复杂机加工零件的三维测量,提供了有效的解决途径。本文主要内容概括如下:(1)深入调研国内外光亮表面的三维形貌测量方法,对现有的方法进行归纳,比较,分析,同时指出该领域仍然存在的难点问题,明确本文的研究内容。(2)针对目前广泛使用的相移法在测量光亮表面时面临图像饱和、互反射和噪声灵敏度高等问题,在深入研究空间编码方法、时间编码方法、相移法等结构光图案编码原理的基础上,提出一种面向光亮表面的快速、鲁棒、高空间分辨率的线移法。为了使生成的图案比正弦条纹图案更可靠,使用格雷码生成正反黑白条纹图案,并像相移法一样进行线移。而在图案解码时,对采集的条纹图像边缘的非线性轮廓做线性插值,通过求解交点得到亚像素精度的边缘坐标,由此达到较高的空间分辨率。(3)测量系统标定是个复杂且耗时的过程,针对基于参考平面的系统标定算法存在约束过强、标定精度不高、可操作性差、需要定制特殊的标定板等问题,提出一种基于特征点映射的系统标定算法。把数字投影仪当作逆向的相机,通过建立相机图像像素和数字投影仪图像像素之间的精确对应关系,将相机拍摄的标定板图像中的特征点坐标映射为数字投影仪图像坐标,从而将数字投影仪参数标定转化为成熟的相机标定,进而将整个结构光三维测量系统的参数标定转化为双目立体视觉系统的参数标定。(4)针对目前在光亮表面三维形貌测量方面存在的自适应测量问题,提出一种自适应、高动态范围的结构光三维测量算法。通过建立光亮表面条纹图案成像的数学模型,分析被测物体表面反射率、表面互反射和环境光照等因素对采集的条纹图像的影响,提出两种生成条纹图案最佳投射灰度值算法,实现自适应调节条纹图案中每个像素点的最佳投射灰度值,以克服由于被测物体表面反射率、表面互反射和环境光照等因素引起的高光和黑暗,从而获得清晰的条纹图像,恢复被测物体的三维形貌。(5)在解相位过程中,针对条纹图像的灰度分布非正弦化引起的相位误差,通过对系统的非线性效应进行建模,在分析系统非线性响应及相位空间分布特征的基础上,提出一种基于平滑样条拟合的相位误差补偿算法。从平面标定板的条纹图像中提取相位误差并构建相位误差查找表,并在后续测量过程中,用于补偿求得的相位。对补偿后仍然存在的残余相位误差,使用平滑样条拟合的方法对相位进一步做光顺处理。(6)在前述理论与技术研究的基础上,设计、开发面向高动态光亮表面测量的结构光三维测量系统。采用Qt应用程序开发框架和OpenCV计算机视觉库开发测量软件,实现系统各项功能模块。最后通过具体的应用实例,验证本系统的检测功能及相关算法的有效性。
[Abstract]:The three-dimensional measurement technology of structured light has been widely used in the fields of industrial detection / quality control, reverse engineering (digitalization of complex free-form surface), object recognition, cultural relics protection, medical and virtual reality. Through this technology, the geometrical information of the workpiece can be obtained accurately, and it is provided for advanced manufacturing, automatic assembly, surface detection and so on. On the other hand, the 3D point cloud of the workpiece is compared with the CAD model or the existing three dimensional data, which can detect the deformation of the workpiece and provide a more complete and more understandable visual analysis method for quality control, stress analysis and collision testing. However, there are still some structural optical three-dimensional measurement techniques in its application. Some problems, such as the limit of ambient light in a certain range when the measurement is required, the measured object must be diffuse surface, and the surface reflectivity varies little. In industrial production, a large number of metal pieces are mainly used for milling. After processing, the surface of the workpiece will become very bright, if the surface is directly on the bright surface. In the measurement, the light after the mirror reflection will be too strong, causing the camera image sensor to saturate and lose the stripe information in the high light region of the stripe image so that it is difficult to carry out the normal measurement. On the other hand, due to the limited dynamic range of the camera, the stripe is too dark to reduce the measurement accuracy for the lower reflectivity area. This paper systematically studies the key technologies of structural light pattern coding, phase error compensation, system calibration and high dynamic range stripe image acquisition, and proposes an adaptive, high dynamic range structure light three-dimensional measurement method, including the line shift coding structured light graph, the system calibration algorithm based on the feature point mapping, and the adaptive algorithm. The optimal projection gray value of the stripe pattern and the phase error compensation algorithm based on the smoothing spline fitting can effectively solve the problem of data loss in the three-dimensional measurement of the bright surface structure light, and provide an effective solution for the three-dimensional measurement of the bright surface of the high dynamic range, especially the three-dimensional measurement of the complex machined parts. The main contents are summarized as follows: (1) in-depth investigation of three dimensional topography measurement methods on the bright surface at home and abroad. The existing methods are summarized, compared and analyzed. At the same time, the difficult problems still exist in this field are pointed out, and the content of this paper is clarified. (2) the current widely used phase shifting method faces the saturation of the image when measuring the bright surface. On the basis of the deep study of the spatial coding method, time coding method and phase shift method, a fast, robust, high spatial resolution line shift method for the bright surface is proposed. In order to make the generated graph more reliable than the sinusoidal pattern, the gray code is used to generate the positive. The anti black and white stripe pattern and line shift like phase shift. While the pattern decode, the nonlinear contour of the fringe image edge is interpolated linearly. By solving the intersection point, the edge coordinates of the sub-pixel precision are obtained. (3) the calibration of the measurement system is a complicated and time-consuming process. The system calibration algorithm of the reference plane has the problems of too strong constraint, low calibration precision, poor operability and special calibration board. A system calibration algorithm based on feature point mapping is proposed. The digital projector is used as a reverse camera, and the accuracy between the pixel of a camera image and the image pixel of a digital projector is established. Corresponding relation, the coordinates of feature points in the camera calibration board image are mapped to the digital projector image coordinates, thus the calibration of the digital projector parameter calibration is converted to the mature camera calibration, and then the parameters calibration of the whole structure optical three-dimensional measurement system is converted to the parameter calibration of the binocular stereo vision system. (4) aiming at the current in the light. An adaptive, high dynamic range of structural light three-dimensional measurement algorithm is proposed in the measurement of bright surface three-dimensional topography. By establishing a mathematical model of the pattern image of the bright surface stripe pattern, the influence of the surface reflectivity, the surface reflection and the ambient light on the Striped image are analyzed. Two kinds of gray value algorithm for the best projection of the pattern pattern are proposed to achieve the best projection gray value of each pixel in the adaptive stripe pattern, so as to overcome the high light and darkness caused by the surface reflectivity, surface reflection and environmental illumination of the measured object, so as to obtain a clear image of the stripe and restore the three of the measured object. (5) (5) in the phase process, the phase error caused by the non sinusoidal distribution of the stripe image is modeled. On the basis of analyzing the nonlinear response of the system and the characteristics of the phase spatial distribution, a phase error compensation algorithm based on the smoothing spline fitting is proposed. The phase error is extracted from the Striped image of the plate and the phase error lookup table is constructed. In the subsequent measurement, the phase error is compensated. The residual phase error still exists after the compensation. The smoothing spline fitting method is used to further smooth the phase. (6) on the basis of the previous theoretical and technical research, the design and development of the phase error are designed and developed. The three-dimensional measurement system of structural light for high dynamic light surface measurement is made by using the Qt application development framework and the OpenCV computer vision library to develop the measurement software to realize the various functional modules of the system. Finally, the detection function and the validity of the related calculation method are verified by specific application examples.
【学位授予单位】：广东工业大学
【学位级别】：博士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】