【作者】 张益昕；

【导师】 张旭苹；

【作者基本信息】 南京大学 ， 微电子学与固体电子学， 2011， 博士

【摘要】 大尺度三维几何尺寸测量技术可以对外形不规则的大型物体进行几何量的精确测量。随着计算机技术与数字图像处理技术的发展,近年来集计算机视觉技术、数字图像处理技术以及光学技术于一身的基于计算机视觉的大尺度三维几何尺寸测量方法已经在质量检测、虚拟现实、逆向工程以及工业制造等领域得到了广泛的应用。但目前的技术在解决室外环境下大量程高速高精度便携式测量应用方面仍存在不足。本文在分析研究现有相关技术与文献的基础上,着重对测量系统的结构设计与软件算法进行研究,得到了一种具有快速、精确、便携以及大量程特点的新型测量方法。论文首先对与成像质量最为相关的照明、镜头以及图像传感器进行了分析和研究,建立了视觉传感器的数学模型并分析了测量误差与传感器结构参数之间的关系,为测量装置的设计提供了依据。论文对计算机视觉测量中的特征提取算法进行了深入地研究,对现有的角点、光条提取算法进行了总结和对比,并分别基于参数自适应以及相位一致性分析给出了角点和光条的提取算法,实现了在现场环境下对图像特征的稳定提取。并设计了自动恢复算法,实现了对离焦模糊图像的恢复。实验结果显示该算法对离焦模糊图像的恢复质量良好,可以有效增强图像特征。视觉传感器参数标定是三维视觉测量中的重要环节,其中对于线结构光传感器的参数标定,传统的标定方法需要昂贵的设备,效率低下,步骤复杂。本文提出了一种便捷的三维视觉系统结构光平面标定方法。该方法仅使用一个简单的二维平面靶标,在保证靶标与结构光相交的前提下,只需要将靶标在摄像机可视范围内自由移动数个位置即可完成标定过程。实验证明该方法效率高,步骤简单,具有很好的通用性。为了解决双目立体匹配中二维搜索空间带来的不稳定性,并进一步提高双目匹配的精度,本文对于基于极线约束与结构光光条约束的亚像素级精度的匹配算法进行了探讨,并提出了一种基于局部曲线差值的高精度匹配算法,实现了对图像特征点的高精度立体匹配。基于本文的理论研究结果研制了一种基于大尺度三维几何尺寸测量的货运列车超限自动检测系统并制作了样机。现场测试结果显示,该系统可以高速准确地完成对货运列车的自动超限检测。其测量的均方误差在10m的测量距离处为3mm,对于单节列车的测量时间不超过30分钟。上海铁路局的科技成果鉴定认为：该项技术达到了路内(国内)先进水平。更多还原

【Abstract】 Large-scale 3D geometric measurement is an effective and flexible method for precise space coordinates measurement when the object’s shape is irregular and the size is relatively large. In recent years, with the evolution in computer technologies and digital image processing, many computer vision based large-scale 3D geometric measurement methods incorporating computer vision technology, digital image processing technology and optics have been developed, which have found wide applications in quality control, virtual reality, reverse engineering and manufacturing. However, these existing methods have limitations when dealing with the portable measurement of large measuring range, high speed and high accuracy under external situation. In this paper, the previously research works and literatures of 3D geometric measurement are analyzed. A novel measurement method with portability, high speed, high accuracy and large measuring range is proposed aims the existing methods’shortages.This dissertation firstly discussed the illumination, lens and image sensor, which are the most relevant factors to the image quality. The mathematical model of the vision sensor as well as the relationship between the sensor’s structure and the measurement error were established. The design rules for measurement equipment were provided.The algorithms for feature extraction have been studied in details. The existing corner and light stripe extraction methods have been summarized and compared. Self-adaptive corner detection method and phase congruency based stripe extraction method were proposed which can work steadily in outdoor situation. A Wiener filter based algorithm for defocus blurred image restoration was presented. The experiment results indicate that the new method effectively reduces the ringing effects and restores the defocus blurred image well.The calibration of sensor is the key technique in 3D vision measurement. For structured light sensor, the traditional methods require expensive equipments and complex procedure. Beside, those methods’efficiency is quite low. In our method, a simple 2D planar target is used. The target can be freely moved to different positions in the visible space of the camera if only the target and the structured light plane can intersect to each other. The experiment results show that the proposed method is a universal method with high efficiency and simple procedure.For stereo vision, the search space for matching one point in the left image to its correspondence in the right image is two dimensional which makes the matching quit fallible. In this paper, epipolar constraint and structured light stripe constraint which can make the corresponding points matching become a one to one situation is used. Although similar constraint has been reported by previous literates, the algorithms in these literates only offer pix level accuracy, which prevent a high overall accuracy of the final measurement. We have proposed a corresponding points matching algorithm based on local curve intersection, which provides sub-pixel level accuracy.Finally, an automatic freight train gauge-exceeding detection method based on large scale 3D geometric measurement was presented. A prototype has been developed and the experimental results have shown that the proposed method can detect freight train gauge-exceeding with high accuracy and speed. The mean square error is about 3mm when the measuring range is 10m. The measurement time for one train is within half an hour. The prototype has passed the acceptance of Shanghai railway administration. The appraisal result proved that our technology has reached the leading domestic level.更多还原