【作者】 黄艳；

【导师】 车仁生；

【作者基本信息】 哈尔滨工业大学 ， 仪器科学与技术， 2012， 博士

【摘要】 针对固定式单摄像机坐标测量系统在大空间测量时测量精度低的问题，本文提出了主动视觉测量的新构想。课题“基于主动视觉的大空间坐标测量关键技术研究”的主要目的是将主动视觉引入现有的基于光学测棒的固定式单摄像机坐标测量系统中，探讨研究一种新的视觉坐标测量方法，以提高大空间测量精度。对这一新的测量系统的关键技术如：主动视觉坐标测量建模、移动光学测棒的捕获、移动光学测棒的跟踪、摄像机相对转台位姿现场校准等进行了深入研究，并提出了解决方案。本文研究了主动视觉坐标测量的系统结构并建立了测量模型。在测量过程中，光学测棒随着被测工件上的被测点改变而移动，系统首先通过摄像机变焦、偏摆和俯仰捕获并跟踪移动中的光学测棒，使不同测距处的光学测棒都以合适的大小成像于CCD中央区域；再根据固定式单摄像机坐标测量模型计算光学测棒测尖也即被测点在此时系统姿态下的三维坐标；最后通过坐标变换将系统各姿态下测得的点的三维坐标拼接在世界坐标下，实现对大型工件的点位坐标测量。对移动中的光学测棒的捕获是指当测棒移动时，摄像机镜头经过变焦和对焦来捕获移动中的光学测棒，使其在像面上仍成大小合适且清晰的图像。本文对变焦镜头做分级变焦，即先根据测距范围和成像目标占像面百分比划分镜头的变焦等级，再估计出变焦镜头的焦距输出曲线，最后在该曲线上快速定位各级焦距，可以获得光学测棒大小合适的图像；为了获得光学测棒上特征光点的清晰图像，本文提出一种将对焦深度法和离焦深度法结合的自动对焦方法，即先用基于退化图像二阶微分自相关的离焦深度法粗估计出正确对焦位置后，再用对点光源的灵敏度最高的Variance函数作为清晰度评价函数的对焦深度法在该位置附近搜索正确对焦位置。实验结果表明在全测量范围内，系统能随着光学测棒的移动完成对其的运动捕获，使其在像面上始终成大小合适且清晰的图像。对移动中的光学测棒跟踪是指当测棒移动时，转台带动摄像机做偏摆和俯仰运动跟踪移动中的光学测棒，使其始终成像于CCD像面的中心区域。为实现快速准确的跟踪，本文提出一种基于图像平面逻辑分区的跟踪规划，即将图像平面划分成理想成像区、跟踪区和搜索区，当特征像点质心在搜索区内时用基于特征像点质心偏差的快速跟踪方法以使光学测棒始终保持在摄像机视场范围内，当特征像点质心在跟踪区内时用基于PBVS的精确跟踪方法以使光学测棒成像于CCD平面的中央区域。实验结果表明该跟踪方法的有效性。摄像机相对转台位姿现场校准是保证主动视觉坐标测量系统测量数据拼接精度的重要环节之一，本文提出一种基于矩阵直积理论的线性算法和基于固定点重复测量的非线性规划相结合的二步校准法。为消除测量噪声的影响，用四元数法正交化所求解的旋转矩阵并计算出新的平移矢量。实验结果表明校准方法具有较高的校准精度。最后，对基于主动视觉的大空间坐标测量系统进行了全面测试，与固定式单摄像机坐标测量系统相比，单点测量重复性在同一水平上，空间尺寸测量精度提高了2-3倍，验证了主动视觉坐标测量系统对扩展测量范围和提高大空间测量精度的有效性。更多还原

【Abstract】 In view of the problem of low accuracy in the measurement of large sacle for staticsolo-vision coordinate measurement system, a new concept of active visionmeasurement has been proposed. The subject “research on the technologies of largescale coordinate measurement based on active vision” is to study a new visioncoordinate measurement system which introduces active vision to solo-visioncoordinate measurement system base on optical probe to improve its large scalemeasurement accuracy. For this novel measurement system, some key problems such asmodeling of active vision coordinate measurement system, acquisition of movingoptical probe, tracking of moving optical probe, calibration of position and attitudebetween camera and rotator have been researched and some solutions have beenproposed.In this paper, we researched the structure of active vision coordinate measurementsystem and the measurement model has been established. First, as optical probe usuallyadapts its position and attitude to the measured point during the whole measurementprocess, acquisition and tracking of the moving optical probe through zooming,rotating and pitching of camera are applied to obtain appreciate size images of opticalprobe in CCD center area; Second, we use solo-vision measurement module to calculatethe optical probe tip coordinate under the current camera coordinate system whichequals the measured point coordinate; At last, coordinate transformation is applied totransform the coordinates gained under every coordinate system to the world coordinatesystem to implement the point coordinate measurement of large scale workpieces.Acquisition of moving optical probe is to capture appropriate size and sharplyfocused images of optical probe at CCD through lens’ zooming and focusing whileoptical probe moving with measured points. Focal length of variable lens is classifiedinto several levels to obtain appropriate size images, that is, the focal length is firstlydivided according to work distance and percentage of image plane, then a fast method isused to estimate the focal length exportation curve, at last locate the specific focallength on this cure. A combined auto focusing approach with depth from focus methodand depth from defocus method is proposed to capture sharply focused images, that is, aDFD method based on autocorrelation of derivative of defocus blured image is used firstly to give a fast but rough estimate of the right focus position, then a DFF methodchosen Variance function, most sensitive to point light source, as criterion function isused to find accurate focus position in the rough field. Experimental results showsystem can acquire moving optical probe and obtain sharply focused images withappropriate size in measurement range.Tracking of moving optical probe is to capture images of optical probe in CCDcentral area through camera’ rotating and pitching while optical probe moving withmeasured points. A new tracking method based on logical partition of image plane isproposed to meet the fast and accurate tracking requirement in active visionmeasurement, that is, image plane is parted into ideal region, tracking region andsearching region. When centroidal of feature image points distributes in searchingregion, a fast tracking method base on the errors of centroidal of feature points is usedto keep optical probe in field of view of camera; When centroidal of feature pointsdistributes in tracking region, an accurate tracking method base on PBVS is used tokeep image of the optical probe in the center of CCD plane. Experimental results showthe effectiveness of this tracking method.Cause the calibration of position and attitude between camera and rotator is the keyto accuracy of measured data combination, this paper presents a valid calibrationmethod consists of linear parameter estimation base on theory of matrix direct productand nonlinear refinement through repeated measurement of fix point. To reduce thenoise corruption, quaternion is applied to orthogonalization the rotation matrix and newtranslation vector is calculated from this orthogonal rotation matrix. Experimentalresults confirm the calibration accuracy.Finally, the collectivity test was done for the large scale coordinate measurementsystem based on active vision, the experimental results show the same repeatabilityfor measurement of single point, but2-3times higher accuracy for measurement ofstandard gauge compared with static solo-vision coordinate system，and approve theeffectiveness of extending measurement range and improving measurement accuracy.更多还原