【作者】 沈满德；

【导师】 陈良益；

【作者基本信息】 中国科学院研究生院（西安光学精密机械研究所） ， 光学工程， 2008， 博士

【摘要】 破片参数测量是武器研发与定型时不可缺少的环节。弹丸爆炸后产生的破片数量大,待测参数多,测量任务非常繁重。传统的人工测量法和光电测量法效率低、人为误差大、自动化程度低,难以满足不断发展的测量需求。计算机视觉检测技术是以现代光学为基础,融光电子学、计算机图像学、信息处理、计算机视觉等科学技术为一体的现代检测技术,具有非接触、全视场、速度快、精度高等优点,具备在线检测、实时分析、实时控制的能力,在军工、机械、电子、农业、医学等领域得到了广泛的应用。计算机视觉检测技术的相关原理与算法可用于破片参数测量系统,从而有效提高系统的测量速度和测量精度。本研究论文根据南京理工大学机械工程学院实验室建设项目“破片参数测量系统的研制”立题,旨在研制一套高速度、高精度的破片参数测量系统。本文通过对计算机视觉检测技术的研究,将该技术中的先进算法与原理用于破片参数测量系统中,以达到系统对测量速度和测量精度的要求。研究过程中参阅了大量文献,对破片参数测量系统的国内外研究状况进行了调研,深入分析了研制该系统的难点,提出了将计算机视觉检测技术的先进算法和原理用于破片参数测量系统的研制思路。概述了计算机视觉检测技术的发展历程、国内外研究现状及发展趋势,明确了检测精度和检测速度两大关键因素对视觉检测系统的重要影响。论文内容主要涉及破片参数测量模型、测量方法、系统组成、精度分析,以及计算机视觉检测中的图像处理、系统标定等关键技术。本文建立了破片迎风面积的测量模型,以正二十面半空间的16个特定方向模拟破片飞行过程中迎风面的取样方向,分别测量破片在这16个方向的投影面积。根据测量模型,构建了基于计算机视觉的破片参数测量系统,该系统由目标照明单元、载物台单元、机械传动单元、光学成像单元、电控单元、软件单元六大模块组成。摄像机在取样方向上采集破片的投影图像,经图像处理求得图像面积,最后根据像素的面积当量将图像面积换算成破片的迎风面积。为了达到单个破片测量时间小于35秒的技术要求,本文采用了四台摄像机并行工作的方式,大大节省了摄像机移动时间。系统软件中采用了多线程技术,使系统主线程与图像处理线程并行,大幅提高了测量速度。数字图像处理是计算机视觉检测的核心内容,对系统的测量精度和测量速度具有决定性的作用。本文着重论述了图像处理中的滤波、阈值分割和边缘检测三方面的内容,提出了适用于破片参数测量系统的图像处理方法。由于景深的限制,系统所采集的图像普遍存在离焦模糊现象,此模糊量会引起较大的面积测量误差。本文提出了两种图像处理方法用来修正离焦图像引起的面积测量误差,一种基于图像梯度最大值,另一种基于边缘检测。实验结果表明,两种修正方法均能达到系统要求的测量精度。在计算机视觉检测中,摄像机标定是一项非常关键的内容。由于摄像机光学系统并不是精确地按小孔成像模型工作,存在着图像畸变。在高精度测量时,需要对图像畸变进行校正。本文分析了镜头畸变产生的原因,介绍了用网格法校正图像畸变的原理。传统摄像机标定方法实时性差,在实际应用中较难实现。本文针对待测物的特点,提出了一种简便、实用而又不失精度的标定方法——特征参照物标定法。该方法不需要标定摄像机内外13个参数,只需标定CCD像元的物理面积当量。实验结果表明,该标定方法具有很高的精度,且不增加图像处理的时间。测量精度是破片参数测量系统最重要的技术指标之一。本文采用不确定度理论对系统精度进行了分析。系统的随机误差主要有系统噪声、测量模型原理误差、旋转台定位误差、离焦成像引起的误差和图像分割误差。对上述各随机误差进行了详细分析,计算了各种误差的不确定度,通过不确定度的合成得到了系统的总不确定度。本文采用实验的方法对系统的测量精度和重复精度进行了验证。按不同的长细比和不同的最大迎风面积设计了四个实验。实验结果表明:平均迎风面积的最大误差小于1.38%,重复测量的最大偏差小于0.03%,测量精度符合并优于技术指标要求。测量结果还表明:破片长细比越小,测量精度越高。最后总结了本课题的研究成果及创新点,并对今后的工作提出了建议。更多还原

【Abstract】 The measurement of fragment parameter is indispensable during the development and confirmation of bomb. The workload of measurement is very heavy, because the number of fragment is very big after explosion and there are too many parameters to measure. Manual method and photoelectric method have some shortcomings, such as inefficiency, big man-made error and low-level automatization, so conventional methods do not meet the requirement in measurement of fragment parameters. Computer vision inspection is a new technology based on modern optics, which syncretizes optical electronics, computer image, information processing, and computer vision. It has many advantages, such as non-contact, big field, high efficiency and high precision. It also has the ability of on-line inspection, real-time analysis and control. It is widely used in military industry, mechanism, electronics, agriculture, medicine and so on. Some principles and algorithms of computer vision inspection can be used to improve the efficiency and accuracy in measure system of fragment parameters.The dissertation is based on the project of the study of fragment parameter measuring system, which is supported by mechanics institute of Nanjin Science & Technology University to develop a fragment parameter measuring system. Some principles and algorithms of computer vision inspection are adopted to improve the efficiency and accuracy of this measuring system.Many papers have been referred in this study. The situation on the research of fragment parameter measurement has been inquired, and the difficulties to realize the project have been analyzed. The idea is put forward that the principles and algorithms used in computer vision inspection will be adopted in fragment parameter measuring system. The history, situation and trend in development of computer vision inspection have been summarized. As critical factors of computer vision inspection, accuracy and efficiency have been discussed. The content of this dissertation is related to measuring model, measuring method, system composition, image processing, camera calibration and accuracy analysis.The measuring model, regular icosahedra, has been established. The system simulates the direction of fragment’s windward using the 16 special orientations of regular icosahedra, and measures the fragments’ projected area in 16 different directions. According to the measuring model, a fragment parameter measuring system based on computer vision inspection has been designed. The system is composed of illumination unit, object stage, mechanical unit, optical unit, electric unit, and software. Image of projected area can be obtained by camera in projected direction, and image area can be calculated throught image processing. According to the equivalent of pixel, the fragment’s windward area can be gotten. The system qualifications require that the time consumption is less than 35s for one fragment measurement. This system uses four cameras to capture image at the same time, so the time consumed in moving the cameras is decreased greatly. Multi-thread technique has been adopted in this system’s software, and the system efficiency has been improved greatly, when main thread and image processing thread works parallel.Image processing is the kernel of computer vision inspection, and has great effect on accuracy and efficiency. In this paper, the basic concept of digital image is introduced, and some methods of image processing, such as image filter, image segmentation and edge detection, are discussed in detail. Based on basic image processing methods, the special image processing method is designed in this measure system. When outside the depth of field, images obtained are usually obscure, so there is big error in the results of area measurement. Two image processing arithmetic’s are developed to amend area error from defocus. The two methods, one based on image gradient and the other based on edge detection, can amend the results to meet the system’s requirements.Camera calibration is a very important step in computer vision inspection. In practice, the camera cannot obey exactly the principle of pinhole imaging, so there is aberration in images. In high accuracy measurement, the aberration should be eliminated. The cause of image aberration is analyzed and the method to correct aberration using grid is introduced. For the requirement of real-time, conventional method of camera calibration is very difficult to realize in practical system. According to the character of images, a new calibration method based on special contrastive object is put forward. In this method, the 13 camera parameters are not need to calculate, and only the area equivalent of pixel is calibrated. The test results indicated that the calibration method is perfect and does not increase image-processing time.Accuracy is the most important technical parameter in this system. The accuracy of this measuring system is analyzed using the theory of uncertainty. In this system, the random errors include noise, principle error of measure model, position error of rotating stage, area error caused by defocus image, and image segmentation error. All errors mentioned above are analyzed thoroughly, and uncertainty of errors is calculated. At last, the whole uncertainty of the system is gotten by synthesis.The accuracy and repeatability of this system are verified. According to the slenderness ratio and the maximum windward area, four experiments are designed. The result shows that the error of average windward area is less than 1.38%, and the maximum deviation of repeatability is less than 0.03%. The accuracy of this system meets the requirement. According to the result, the more slender the fragment is, the less accurate the result is.At the end of the dissertation, the research results and original discoveries are presented, and the succeeding research is suggested.更多还原