【作者】 史艳琼；

【导师】 卢荣胜；

【作者基本信息】 合肥工业大学 ， 精密仪器与机械， 2012， 博士

【摘要】 变焦(变倍率)光学成像系统是指被摄物距在沿光轴方向发生变化时,为了得到合适大小的图像,光学系统的焦距可以在一定范围内连续调整的光学系统。变焦光学成像系统目前在摄影、摄像、安全监控、视觉跟踪、精密测量等领域得到广泛应用。民用摄影领域中,变焦距镜头已经代替固定焦距镜头,成为照相机的重要组成部分。在全景成像时,让变焦镜头绕着透视投影中心旋转,拍摄多幅图像,采用图像拼接技术可获得更大画幅的全景图像。在视觉跟踪与导航领域,使用方位角、俯仰角和倍率可以调整的Pan-Tilt-Zoom (PTZ)摄像机或相机,在被跟踪目标距离发生变化时,能够保持跟踪得到的图像大小不变,同时保持较高的影像分辨率,还可能获得目标的姿态和位置信息。因此,研究变焦光学系统成像过程中的相关技术与科学问题,具有重要的学术研究与应用价值。对于由固定焦距组成的定焦光学系统,视觉模型参数是固定的常量,模型的定标只需要我们对系统进行测量和估计模型中常数项的值。但对于变焦镜头组成的光学成像系统,这些参数随着不同镜头结构设置而变化,因此使用变焦光学系统进行跟踪、定位与测量,在技术上具有巨大的挑战。本论文在国家863项目课题的支持下,针对变焦光学成像系统在精密跟踪、定位与测量中存在的一些科学与技术问题,开展分析与研究。论文主要研究工作和取得的研究成果如下：(1)针对变焦光学系统结构参数随着倍率、对焦和光阑孔径大小的调整而发生变化的特点,对变焦光学系统的小孔模型和厚透镜模型进行了详细的分析与讨论,研究结果表明厚透镜视觉模型具有更高的建模精度,更适合变焦光学成像系统的建模。(2)从理论分析和实验验证两个方面出发,阐明了变焦光学系统小孔透视模型透视中心的物理位置,分析与研究了变焦成像光学系统影像中心的变化规律。研究结果表明,变焦光学系统的透视中心在变焦镜头的入瞳位置；在变焦光学系统视觉透视建模时,影像中心位置应该用透视中心到像面垂线的交点来定义；为了获得理想的模型精度,变焦镜头的入瞳位置在变焦过程中应保持恒定不变。这些结论为建立高精度的变焦光学系统视觉模型和设计能够用于精密跟踪、定位与测量的变焦光学系统,提供了理论支持。(3)研究了PTZ摄像机通过改变水平旋转角、俯仰角、倍率和对焦等参数,实现跟踪、定位与测量物体的方法,建立了包含全局坐标系、水平和俯仰旋转云台坐标系、相机坐标系、实时变化的相机内部参数的三维视觉模型数学表达式。(4)研究了一种基于PTZ摄像机的视觉跟踪6D测量技术及系统。该变焦视觉跟踪6D测量系统主要由自变焦视觉跟踪摄像机、彩色伪随机编码六自由度测头和计算机组成,利用测头编码标准球上的彩色伪随机编码在摄像机成像面上的位置变化来确定靶标的姿态,从而实现大尺寸空间六自由度的测量。在组合测量、大型设备安装定位中具有广阔的应用前景。(5)研究了变焦视觉跟踪图像中基本几何特征——角点、直线和椭圆等基本几何特征的识别方法与算法。针对经典Hough变换算法直线检测技术的不足,提出了改进的算法,提高了检测准确度的同时降低了运算时间；研究了两种基于计算边缘曲率的新的角点检测算法——小波变换检测法与轮廓多边形近似检测法；研究了在强噪声环境下提高椭圆检测正确率的迭代Hough变换算法；给出了一种新的椭圆检测算法——改进的曲线弧分割检测法,通过实验验证了此方法可用于解决Hough变换算法与最小二乘算法的遗留问题。(6)总结了像素级和亚像素级边缘检测技术,提出了两种改进的亚像素边缘检测算法,一种是基于高斯曲面拟合亚像素边缘检测算法,另外一种是基于径向基函数插值亚像素边缘检测算法。这两种算法的基本原理是首先选用Canny算子对图像进行边缘粗定位,然后在粗定位的边缘基础上给定一个窗口,对窗口内进行精细定位。实验结果表明,这两种算法均可达到0.1个像素的精度,分辨率为0.03个像素。(7)基于研制的PTZ摄像机的视觉跟踪6D测量实验系统,对本文提出的一些观点、算法进行了实验验证,对PTZ变焦跟踪实验系统的性能进行了测试,总结了PTZ变焦跟踪视觉测量的特点和应用前景。论文最后还对未来的工作重点进行了简要说明。更多还原

【Abstract】 Zoom optical imaging systems like zoom cameras, photogrammetry cameras, PTZ cameras and optical theodolites are optical systems with their focal lengths can be continuously adjusted in certain extent for acquiring appropriate images of objects when the objects distance changes along the optical axis of the systems. Currently zoom optical imaging systems have been validly used in photography, photogrammetry, safety monitoring, vision tracking and high precision measurement areas.In the civilian photography, vari-focus lens have been taken place of the well-known fixed-focus lens and become the important part of the photographic cameras. In panorama photography, multiple single-images can be stitched into a large full view of panorama picture through rotating vari-focus lens around perspective projection center or called as the no-prarllax point (NPP). In vision tracking and navigation area, Pan-Tilt-Zoom(PTZ)cameras, whose azimuth angle, tilt angle and zoom ratio are adjustable, are used to track target with keeping the images size and imaging resolution constant, and to obtain the target’s orientation and positions. It is apparent that the research on the theoretic and technical issues of the vari-focus optical system and their imaging process has important academic and application values.In machine vision, the visual model parameters of fixed-focus optical image system are fixed in applications. The requirements for the model parameters’calibrations are to measure and estimate the values of the corresponding constant terms. For zoom optical image systems consisting of vari-focus lenses, however, the model parameters vary with the configuration setting adjust, and it is a big challenge to use zoom optical image system to implement tracking, position and measurement. The dissertation, which are supported by national863program, are concentrated on investigating some key scientific and technological topics in application of zoom optical imaging system for high precision tracking, positioning and measurement. The important research work and achievements in the dissertation are listed as follows:(1)The pinhole model and thick lens models of zoom optical imaging systems are investigated and summarized to understand the change rule of model parameters of the zoom optical imaging systems affected by zoom (focal length), focus and iris. The research results shows that thick model is better to model the zoom optical imaging systems with higher accuracy。(2) The physical positions of perspective projection center of the pinhole model of zoom optical imaging systems, and image centers are studied through theoretical analysis and experimental verification. The research results demonstrate that the perspective projection center is right at the intersection point between the entrance pupil of vari-focus lens and the optical axis in modeling the zoom optical imaging system. The image center is reasonable to be defined by the image point on the image plane intersecting by the perpendicular line from perspective center to image plane. To obtain high accuracy visual model, the entrance pupil of vari-focus lens are needed to keep constant in zooming. The conclusions offer theoretical supports for setting up a high precision zoom optical imaging system and modeling the system in high precision tracking, positioning and measurement.(3) The methods of tracking, positioning and measuring an object using PTZ cameras with change of their zoom (focal length), focus and iris are researched. The research involves establishing the numerical expression of the3D vision model with internal and external parameters adjustable and the model coordinate systems including global coordinates, PTZ coordinates about horizontal and tilt rotation, camera coordinates.(4) A six-dimensional (6D) measurement techniques and system based on PTZ cameras have been developed. A zoom tracking visual measurement setup has been built up for research on the topics of the zoom tracking, positioning and measurement. The setup consists of an auto-focus zoom tracking camera, a6freedoms probe with a standard sphere, whose surface is covered by pseudorandom color codes, and a computer. The pseudorandom color codes are distributed on the sphere with color arrays of holes, where color LEDs lighter are assembled. The hole arrays with the known center coordinates are used as a3D calibration target for the tracking system parameters self-calibration. Each hole position is recognized using the color emitted by its LED. The6D information of the probe is obtained based on the coordinates of the color array in the system coordinates. The tracking measurement setup has good potential application prospects in combination measurements and large equipment assembly process.(5)The methods and algorithms of recognizing the basic geometric features, such corner, line, ellipse and etc. in zoom vision tracking images are studied. In view of the drawbacks of the classic Hough Transform algorithm in line detection, an improved algorithm, which can increases detection accuracy and reduce computing time, is developed. Two new corner detection algorithms based on calculating edge curvature are studied. One is based on wavelet transformation and another on polygonal approximation detection. An iterative random Hough transform which can enhance ellipse detection accuracy under strong noise environment is investigated. A new ellipse detection algorithm called curve segmentation detection algorithm is proposed. The algorithm has been proved to be of effectiveness by experimental verification and be able to solve the remained problems which cannot be resolved by classical Hough Transform and Least Square method respectively.(6) The pixel edge detection and sub-pixel edge detection technologies are summarized and two modified sub-pixel edge detection algorithms are proposed. One is based on gauss surface fitting technique, another based on radial basis functions interpolation techniques. Basic principles of the two algorithms can be briefly described as follows. Firstly the Canny operator is used to detect the coarse position of an edge, and then the sub-pixel edge detection of the edge is implemented within the small area of the coarse position. Experimental results show that two algorithms can reach0.1pixels accuracy and0.03pixels resolution separately.(7) A6D measurement experimental setup is built up based on the tracking, positioning and measurement techniques of PTZ cameras. The setup then is used to investigate and verify the ideas and algorithms proposed in the dissertation. The performance of the setup has also been tested. The characteristics and application prospects have been summarized.At the end of the dissertation, the future research topics are briefly introduced.更多还原