【作者】 郑毅；

【导师】 刘上乾；

【作者基本信息】 西安电子科技大学 ， 电子科学与技术， 2008， 博士

【摘要】 管道静态参数测量是特种管道设计、研究和使用中必需且重要的技术环节。随着特种管道研制、生产水平和使用要求的不断提高,原来基于光学望远系统的人工测量、人工处理数据的测量方法已经不能满足需要,必须建立高精度、高效率和高度自动化的测量系统。在对特种管道静态参数的测量原理、测量方法和测量装备的性能及其局限性进行深入研究的基础上,我们设计并实现了一种以爬行器运动控制技术、照明技术、图像预处理方法和高精度图像测量技术为核心的管道静态参数测量系统方案。这些技术和方法是测量特种管道诸多静态参数所需的共性技术,因此论文仅以特种管道静态参数之一的弯曲度参数的测量为例,对系统涉及到的上述关键技术问题进行了深入地研究,主要包括以下五个方面的内容:1.提出了爬行器运动控制系统进深运动的动力学模型,以及均匀照明模型根据系统实际工作情况,提出了爬行器运动控制系统进深运动的动力学模型,并深入研究了负荷大小、载体转动惯量与电机扭矩的大小及其变化关系。同时还研究了爬行器运动控制系统进深定位精度的影响因素。根据照度均匀设计原则,提出了均匀照明模型,并针对管道内部空间狭小、测量系统结构尺寸小、成像光学系统物距短、视场角大等特点,研究了在这种特殊环境下实现清晰成像所必需的照度分布条件及实现技术途径。2.提出了根据局部梯度模值判别靶标图像中各像素的噪声类型,并使用复对数Gabor小波滤波器去除相应噪声的靶标图像噪声抑制方法通过在降噪之前引入局部梯度模作为阈值,判断图像中每一个像素所受噪声干扰的类型,根据不同的噪声类型使用不同的降噪方法。对受到椒盐噪声干扰的像素,使用中值滤波降噪。然后,对去除椒盐噪声的图像,利用复对数Gabor小波提取各像素的相位信息和幅度信息,确定最小尺度滤波器对的噪声幅度分布的估计值,从而自动地确定各个尺度上的噪声幅度分布的估计值和噪声萎缩阈值,去除靶标图像中的高斯噪声。3.针对亮度不均匀的靶标图像,提出了利用变分的自适应阈值曲面图像分割算法和基于最大模糊熵的图像阈值分割快速算法利用变分的自适应阈值曲面图像分割算法可以从非均匀亮度图像中将靶标图案有效地分割出来。但由于偏微分方程迭代运算耗时太大,使这种方法不能满足测量系统的实时性要求,因此我们又提出了基于最大模糊熵的图像阈值分割快速算法,使用了遗传算法和模拟退火算法来加快算法的收敛速度。4.提出了RBF网络校正法和基于直线特征的非线性畸变校正法,校正靶标图像的几何畸变基于RBF网络的靶标图像畸变校正方法能够克服多项式变形方法中的方程组奇异解的问题。但这种方法仍然需要准确获得控制点的畸变位置坐标和无畸变位置坐标,否则畸变校正的效果不理想。因此,我们又提出了一种基于直线特征的网格靶标图案的畸变校正方法。这种方法不再需要获取畸变图像与校正图像的控制点对,只要求成像视场中存在不通过视场中心的且具有一定长度的直线特征的空间对象,就可以求解畸变系数,完成畸变校正。5.提出了一种将LoG算子和Zernike矩算子相结合的两步高精度边缘检测方法先使用LoG算子将目标(靶标图案和激光光斑)边缘定位到像素级精度,然后对这些边缘像素的邻域使用Zernike矩算子进行处理,得到目标边缘的亚像素位置。对光斑边缘的亚像素位置坐标使用最小二乘法拟合,可以得到光斑拟合圆,并计算出光斑形心位置。同时,对影响管道静态参数测量系统测量精度的误差源也进行了深入地分析,并给出了相应的解决方法,而且对进一步的研究工作进行了展望。更多还原

【Abstract】 Measurement of tube static parameters is a prerequisite and important technique process in designing, studying and use of the special type of tube. The laggard measurement method based on optics glass and data processing artificially cannot satisfy new requirements that following progression of the development, manufacture level and use of the special type of tube. A high efficiency, precision and automatism measurement system need be constructed.Based on thorough study of the measuring principle, measuring method and performance and limitation of the special type of tube, a project to the tube static parameters measurement system with the core of crawler control, illumination, image pretreatment and image measurement at high accuracy has been designed and realized. These technologies are absolutely necessary for all the static parameters measurement, so key technologies for camber measurement are only studied thoroughly in this dissertation. The following is the main contents that studied in the dissertation.1. A kinetic model on the movement of the crawler control system and an even illumination model are proposedAccording to the operation demands of the crawler, a kinetic model of the crawler entrance motion is proposed. And the load, the carrier’s moment of inertia and motor torque are studied thoroughly. At the same time, the factors effecting motion precision have also been studied.According to even illumination method, an even illumination model is proposed. In view of narrowness of interior space, small size of the measurement system, short object distance and large field of view of imaging optical system, the condition illuminating evenly and realization method are studied in the special environment.2. Many kinds of noise can be distinguished by the local gradient module of pixels, and the corresponding noise can be denoised with complex-valued log Gabor wavelet filterThe local gradient module of pixels is regarded as the threshold value, by which many kinds of noise can be distinguished. Then the corresponding method can be chosen to remove noise. The median filter is worked on pixels polluted by salt and pepper noise. With the use of complex-valued log Gabor wavelet, phase information and amplitude information of a pixel can be obtained in the partially denoised image. And an estimate of noise amplitude distribution for the smallest scale filter pair can be got. From the statistics of the smallest scale filter response, estimates of the noise amplitude distributions at all the other scales can be obtained, and shrinkage thresholds can be set automatically. Thus the noisy image can be denoised completely.3. In view of target images with nonuniform illumination, an adaptive threshold surface method based on variation and a fast image segmentation method based on maximum fuzzy entropy are proposedAn adaptive threshold surface method based on variation can segment target images accurately. Because of time-consuming iterative operations of partial differential equations, the adaptive threshold surface method based on variation cannot satisfy adequately the requirement of real-time image segmentation. Subsequently, a fast algorithm of image segmentation based on maximum fuzzy entropy is proposed to satisfy the requirement of real-time segmentation. Genetic algorithm and simulated annealing algorithm are adopted to not only generate better solutions but also accelerate the speed of convergence in the proposed fast algorithm.4. Nonlinear distortion correction methods based on Radial Basis Function neural network and linear characteristic are proposed to correct target images A nonlinear distortion correction method based on Radial Basis Function neural network can keep away from the difficult problem of singular solutions of equations in the polynomial warping method. However, the method requires the accurate determination of point correspondences between a scene and an image of that scene, or else corrected images are unsatisfying. Thus a nonlinear distortion correction method based on linear characteristic in the target pattern is proposed. The method does not require the determination of point correspondence between a scene and an image of that scene. The method simply requires images of known linear segments, which do not pass through the center of field of view and are long enough to be calculated in the distortion correction operation.5. A two-step edge detection method at high accuracy combined with LoG operator and Zernike moment-based operators is proposedEdges of the target pattern and the laser spot are reached at pixel accuracy by using LoG operator, then they are reached at sub-pixel accuracy by using Zernike moment-based operators in the extending concourse of edge pixels. The edge pixels coordinates at sub-pixel accuracy are used to fit spot circle and calculate the centroid of spot in least square estimation method.At the same time, error sources affecting measurement accuracy in the tube static parameters measurement system have been analyzed thoroughly, and the corresponding methods have been proposed to overcome the deficiencies. And we look forward to more researches on the system.更多还原