【作者】 赵阳；

【导师】 曲兴华；

【作者基本信息】 天津大学 ， 测试计量技术及仪器， 2011， 博士

【摘要】 视觉测量目前已成为一项应用非常广泛的精密测量技术,随着计算机技术的延伸和拓展,视觉测量技术也受到了更广泛的关注。本文提出了基于平行光照明的三目视觉测量系统方案,通过选择平行光源照明方式,可改变目标在各位置相机中的显现方式和显现力强度,利用三相机图像信息的差异性实现互补或者差分融合,能够提高前景目标的显现力,并去除掉背景噪声或非测量目标的干扰影响。目标前景的对比度提高、显现力增强,对于分离前后背景信息,提高目标特征边缘饱和度、均匀性都是非常有帮助的,为实现更高精度的特征形貌检测或者轮廓尺寸测量打下了基础。本文研究的主要内容包括:1.采用了基于平行光照明的三目视觉测量系统方案,讨论了该系统的设计原理及意义。完成了硬件系统的搭建,主要包括:照明光源系统、视频采集系统、多自由度可调节平台等部分。通过平行光照明系统的光学参数、机械结构和电气控制系统的设计,研制了光源系统。搭建了可以灵活调节水平角度和三个自由度位移量的平台系统。2.对系统的图像匹配特征点检测算法进行了研究。为了获取多图像中的匹配特征点,针对匹配图像中常见的角点、圆和椭圆特征分别提出了智能检测方案:基于harris原理的自动角点检测算法,基于hough变换的圆特征自动检测算法,结合hough变换和椭圆拟合原理的椭圆特征自动检测算法。通过引入不同的实验对象,分析了目标偏转角度对于角点提取的影响。以提取的圆特征和椭圆特征中心的行、列直线度来评价两种特征检测算法的特征提取精度。对比分析了圆特征自动检测算法及椭圆拟合算法对椭圆特征的处理结果。3.对系统的图像配准和相机标定技术进行了深入分析。采用了基于特征点的2D图像配准方案,并提出了DLT和Ransac相结合的图像2D射影变换关系计算方法。设计实验分析了在高斯噪声和野值点对的影响下,DLT和Ransac组合算法的运算精度、时间复杂度等情况,并通过对不同目标的实验处理,分析算法的实际匹配效果和精度。由浅入深的分析了相机几何模型,实验中提出了三种相机内参标定方案,并结合不同实验对象,分析了不同标定条件对相机内参数计算的准确性和稳定性的影响。4.采用了基于前后景采样点对比度和区域灰度评价的图像融合系数优化方法,分析了不同系数的融合结果的特点,实现了金属工件缺陷及表面细微凹凸、起伏变化形貌的检测。分别以金属工件和量块为对象,对比单目相机系统讨论了本系统在目标特征轮廓尺寸测量方面的特点。为了实现目标不同信息(比如内、外边缘和缺陷边缘)分离显示,提出了基于Radon一维定向投影的目标分类信息提取方法。通过获取不同分类信息的位置特点,使信息分离,便于对目标各信息的轮廓尺寸进行分析。5.以两种高漫射率目标为对象,分析了高漫射率目标的三相机成像效果,并提出了基于序列差分的多图像融合策略,比较两种应用目标的采样点对比度走势,分析了它们的漫射率效果,提出了系统在目标反射、漫射效果方面的评价方法。更多还原

【Abstract】 Visual measurement is widely used in precision measurement technology. With the extension and expansion of computer technology, visual measurement technique has also received more attention. The thesis has proposed a series of relevant key topics about trinocular vision measurement system based on the illumination of parallel light. The manifestation method and manifestation intensity of the cameras at each site could vary for the choice of parallel light as the source. Through the processing method of complementarity or difference fusion of the image differences obtained by multiple cameras, the manifestation intensity of the foreground could be highlighted and the background noises or the interference effect of the non-measurement object could be removed. It could be very useful to separate the foreground information from the back ground to improve the saturation and uniformity of the feature edges. So this method could achieve higher precision in characteristic morphology detection and coutour dimension measurement than traditional signal-camera system.The main contributions of the thesis include:1. A trinocular vision measurement system based on the illumination of parallel light is proposed and the design principles and significance are discussed. The hardware of the system is designed and built, mainly including light source system, video capture system, adjustable platform of multi-degree of freedom and etc. With the design of the optical parameters、mechanical structures and electrical control system of the parallel light, a light source system is developed. Additionally, a platform system is built up, which could make the devices flexibly rotate at horizontal plane, and move along the directions of three degrees of freedom.2. The algorithm for detecting matching feature points of images in the trinocuar vision system are studied. In order to obtain matching feature points in three images, we propose some intelligent detecting methods for the features of corner point, circle, ellipse feature, including automatic corner points detection algorithm based on harris principle, automatic circle features detection algorithm based on hough transformation principle, automatic ellipse features detection algorithms based on the combination of the harris and hough transformation principles. The influences of target deflection angle on corner extraction are analyzed through different experiment targets. The row、column straightness obtained from the center points of circle or ellipse features is used to judge the extraction accuracy of the detection algorithms. The methods of detecting ellipse features with Both of the circle feature detection algorithms and ellipse fitting algorithms are comparative analyzed.3. The image registration and the calibration technique of the system has been deeply analyzed. 2D image matching method based on feature points is used, and a calculation method of 2D projective transformation relationship based on DLT and Ransac is proposed. Experiments are designed to analyze the influence of Gauss noise and wild point pairs on precision and time complexity when the DLT and Ransac combined method is used. Moreover, the actual matching effect and precision of the calculation method are analyzed through experiments processing on different objects. The geometric models of the camera system are analyzed and three kinds of camera internal parameters calibration schemes are proposed. Combined with different objects, we analyze the influences of different calibrate condition on the accuracy and stability of camera internal parameters calculation.4. A image-fusion coefficient optimization method is proposed which is based on the judgment methods of both the contrast ratio of the sampling points and the grey region. The characteristics of image fusion results with different coefficients are analyzed. With different fusion schemes, information, including defects detection, micr-concave-convex or fluctuant changes on the surface of workpiece, to be detected. The metal workpiece and gauge block are used as targets separately. It is discussed that the trinocuar vision system is much better in the measurement of contour dimension than the single vision system. In order to achieve the separation of the different information( such as the inner、outer edge and defect edge ), we proposed an classified information extraction method based on Radon, which is a one-dimensional directionally projection method. With the obtained positions of the different classified information, we could separate them from the image to conveniently analyze the contour dimensions for each information.5. The trinocuar vision imaging effect of the high diffusion target is analyzed through experiments of two high diffusion targets. And three-image fusion strategy based on serial image difference is proposed. Comparing the contrast ratio trend of the sample point in two targets, the different diffusion effects of these two targets can be understood. Then a judgment method is proposed about the reflection、diffusion effect of different material.更多还原