【作者】 王宪；

【导师】 谭建平；

【作者基本信息】 中南大学 ， 机械工程， 2013， 博士

【摘要】 以计算机视觉为基础的测量技术,具有非接触、灵活性强、集成性高等显著优势,在工业测试领域有着广泛的应用前景。在极端恶劣的工业环境中,振动、噪声、水雾、粉尘等干扰都会对视觉测量造成不利影响,稳定性和精度难以保证。当系统长期在恶劣的条件下工作时,上述干扰因素还将造成测量装置性能下降,进一步降低检测精度,甚至造成系统失效,而作为保证系统稳定工作的重要途径——系统维护在恶劣工业环境下难以实施,需尽量减少,现有视觉测量技术难以保证强干扰条件下检测的准确性和长期可靠性。研究能稳定准确应用于强干扰工业条件下的视觉测量技术是一项具有实际意义和挑战性的工作。本文在国家科技重大专项“两万吨难变形合金卧式挤压机”(项目号：2009ZX04005-031)、西南铝业集团技术改造项目“125MN挤压机技术改造”及航空航天九院国营江北机械厂预研项目“基于机器视觉的柔性接头摆心测试系统”支持下,采用理论分析和实验研究相结合的方法从特征提取、图像增强、精度校准及数据处理四方面入手,对强干扰条件下精密视觉测量技术进行了较系统和深入的研究,并集成上述研究成果,实现了实际工业测量系统的软、硬件设计和完善。论文的主要工作和创新成果可归纳如下：1.结合工业环境的特点,较系统的分析了常用的图像分割算法及圆特征提取算法的适用范围、优势和局限。提出了一种基于区域和边缘准确定位的圆特征精确提取方法。该方法以在干扰条件下表现优良的Canny边缘检测器、Hough变换和和曲线拟合特征提取技术为基础,可在较强干扰条件下快速实现较高的特征提取精度,还可用于复杂背景下多特征检测。2.讨论了常见噪声抑制及图像增强方法的优势及局限。针对极端恶劣条件下极低质量检测图像最普遍存在的问题,结合距离变换和连通分量分析理论,提出了一种应用像素邻接特性分析的激光边缘图像修复方法。该方法仅通过对同一张标号图像的分析即可实现同时存在大断裂和大噪声的极低质量工业检测图像边缘有效修复,引入的中心定位误差稳定的保持在较低的水平,对以圆(椭圆)为检测特征的测量系统具有普适性。3.分析了透视原理下检测系统线性及引入非线性畸变的成像模型,在现有方法基础上,针对工业在线检测的特点和要求,提出了一种应用直线畸变方程的工业测量系统精度校准及精度维护方法。该方法改传统基于优化迭代求解系统参数过程为数据拟合式,降低了对环境噪声的敏感程度；校准过程分图像畸变校正和位置误差校正两步,对特征点的布置方式进行了优化,过程简便。给出了新方法的主动视觉实施实例,测试表明,该方法精度较高且易于实施,为用于工业在线监测的测量装置精度校准及精度维护提供了一种有效的解决方案。4.为进一步提高视觉测量系统在工业检测中的性能,分析基于贝叶斯理论的一系列滤波方法,结合工业在线检测的特点,讨论了各自的优势和局限,选择Sage-Husa自适应Kalman滤波及其改进算法作为研究重点,提出了Sage-Husa自适应滤波算法在测量系统误差抑制和特征提取两个层面上的应用。5.将本文的部分研究成果作为系统升级应用到了125MN挤压机活动部件中心在线监测系统中；集成上述研究成果及电子测量技术最新进展,提出了大型挤压机活动部件五自由度监测系统方案和基于机器视觉的柔性接头摆心测试方案,并对柔性接头摆心测试方案进行了现场测试。更多还原

【Abstract】 The measurement technology based on computer vision has many significant advantages such as non-contact, flexibility and high integration, so it has a widespread application prospect in industrial testing. In extreme industrial environments, vibration, noise, mist, dust and other disturbances could have an adverse impact on vision measurement in the aspects of stability and accuracy. If a system always works under harsh conditions, the interference factors above will also cause the measurement device to deliver poor performance, to further reduce the detection accuracy and even to suffer system failure. System maintenance, as an important way to ensure system stability is to be minimized because it is difficult to implement in harsh industrial environment. It is difficult for existing vision measurement technology to guarantee accuracy and long-term reliability of detection under the condition of strong interference. Research of the vision measurement technology stable and accurate under strong interference conditions is a realistic and challenging work.Funded by the national science and technology major project,200MN Horizontal Extruder for Super Alloy (2009ZX04005-031), the technical transformation project of Southwest aluminum group, Technological Transformation of125mn Extruder, and the pre-research project of the state-operated Jiang-bei machinery factory under the No. Nine institute of aeronautics and astronautics, Computer Vision-based Swing Center Testing System for Flexible Joint, this paper present the systematical and intensive research of precision vision measurement technology under strong interference conditions from four aspects of feature extraction, image enhancement, precision calibration and data processing by combining theoretical analysis and experimental research. The hardware and software design and improved on the real industrial measuring system were achieved by integrating the above research results. The major work and innovations can be summarized as follows:1. The applicability scope, advantages and limitations of common image segmentation and circular feature extraction algorithm were systematically analyzed in light of the characteristics of industrial environment; and a fast precise circular feature extraction method based on regional and edge accurate positioning. This method based on Canny edge detector, Hough transform and curve fitting feature extraction technology gives excellent performance under noisy conditions and realizes high-precision feature extraction rapid under strong jamming. It can be used for multi-feature detection under complex background also.2. The advantages and limitations of common noise suppression and image enhancement method were discussed. For the most common problem in detection of very low quality images under the extreme conditions, a laser edge image inpainting method based on pixel adjacency analysis was presented combining distance transform and connected component analysis theory. This method can effectively repair larger edge gap and remove larger noises through the analysis of one single label image while the value of introduced error of centering measurement is steadily kept low. The method could also be applied to the measurement system whose detection feature is circle or ellipse.3. A linear imaging model and an introduced non-linear distortion imaging model by the principle of perspective were analyzed. To meet the requirement of industrial online measurement and in light of its characteristic, a method for accuracy calibration and maintenance of the industrial measurement system based on straight line distortion equation was presented. This method changes the parameter process of traditional optimized iterative solution system to data fitting, reducing the sensitivity to ambient noise; the calibration process consists of two steps:image distortion correction and position error correction, the layout of feature point, after optimization, become simple. An active vision implementation example by the new method was presented. The experimental results indicated that this method is high precise and easily applicable. It is an effective solution for accuracy calibration and maintenance of the industrial measurement system.4. In order to further improve the visual measuring system performance in the industrial test, some work was done to analyze a series of filter methods based on Bayesian theory, discuss the advantages and limitations of each method in light of the characteristics of industrial online detection. Sage-Husa adaptive Kalman filter and its improved algorithm werw chosen as the focus and the application of Sage-Husa adaptive filtering algorithm in system error suppression and feature extraction proposed.5. Part of research results in this paper was applied as system upgrade to the125MN extruder moving parts center online monitoring system. A real-time monitoring method for five-degrees-of-freedom of the large extruder’s moving parts and a computer vision-based swing center testing method for flexible joint were proposed integrating the above research results and the latest progress of electronic measurement technology. A field test of the swing center testing method was conducted.更多还原