【作者】 王磊；

【导师】 高隽； Peter Ott；

【作者基本信息】 合肥工业大学 ， 信号与信息处理， 2007， 博士

【摘要】 信息获取是信息科学的重要分支，是传统传感技术与其他多学科的发展与交叉融合的产物，表现为信息获取的高精度、高速度、集成化、智能化等特性。三维信息的获取是信息获取的重要组成部分，研究如何获取物体在三维空间中表现出的几何结构和尺度信息，主要是在传统的二维投影尺寸上增加深度信息，更全面真实的表示被测物体。随着科技水平的不断发展，三维信息的获取在科学研究和工业生产中体现出越来越重要的地位。本文以三维信息获取技术为主线，针对集成视觉系统的旋转对称三角位移传感器(RST Rotationaly Symmetric Triangulation)，从信息获取的传感器融合、有效性、物理极限和测量效率等角度，研究了高分辨率三维信息获取技术。主要工作包括：1研究了旋转对称型激光三角位移传感器(RST)与视觉测量系统的集成。给出了传感器的原理与设计，实现了多传感器在物理层上的融合。2研究了RST传感器在安装和调整时引入的系统误差与补偿技术。提出了该传感器的几何光学测量模型；重点研究了由于投射激光偏离旋转轴和图像传感器的倾斜等因素造成系统旋转对称性下降而引起的误差因素；针对旋转对称性下降情况下的圆环缺失，提出了基于神经网络的误差补偿方法。给出了该传感器实验样机位移测量的不确定度。3研究了RST位移测量的不确定度极限。重点研究了RST中的激光散斑现象，从散斑统计学的角度，首次推导了激光散斑引起的旋转对称三角法位移测量不确定度的极限，结果表明在光学粗糙表面上，该极限由投射激光波长，光学系统入瞳对投射光斑所张的立体角，以及光学系统主光轴与旋转轴的夹角确定；这一结论同样适用于其他变形成像(anamorphic)的激光三角法。4基于RST的工作方式，研究了利用集成视觉系统的扫描移动来获得被测物体的低分辨率三维信息的方法。针对近景摄影测量中被测物缺乏足够的纹理细节问题，提出了一种基于对称多基线图像序列的匹配方法；利用亚象素匹配提高其精度，并获得一个连续的三维边缘；仅针对物体的边缘特征进行匹配，从而提高了三维信息获取的速度和鲁棒性。5研究了三维特征指导下的快速高分辨率测量。针对RST传感器获取的高分辨率三维信息是限制在与扫描平面垂直的方向上，提出了将三维轮廓投影到扫描平面上进行表面分块处理和特征提取的新方法；使用该方法进行了三维框架的分割；提出了二维边缘驱动的和三维分块表面驱动的两种测量路径规划的方法；最后将高分辨率距离数据与三维边缘融合，提高了RST传感器的测量效率，同时可以获得自然二次曲面的参数表达。更多还原

【Abstract】 Information acquisition is an important branch of the information scienc. It’s a result from the progress and intersection of traditional sensing technology and other information technologies, and is charactered by the high accuracy, high speed, integration and smartness in the process of information acquisition. 3-dimentional information acquisition is an active field in information acquisition, which studies methods of acquiring the geometrical structure and size of 3D objects. Normally it means getting the range data in addition to traditional 2D projective scale. With the development of then siecne and technology, the 3D imformation acqurisition is more and more important in research and industrial field.In this thesis, based on a rotationally symmetric triangulation sensor with integrated vision system, some subjects in 3D information acquisition are investigated, including the sensor fusion, the validity, the uncertainty limit and the speed. Main works in this thesis include:(1) The integration of rotationally symmetric triangulation (RST) sensor and the vision system is studied. The mockup of the sensor system is built and the fusion in sensors’ physical layer is realized.(2) The system error factors are investigated. The geometric optical model and the error model of RST sensor are studied. More attention is paid to errors caused by asymmetry which is resulted from eccentric laser illumination and tilted image sensor. It is pointed out that if some part of the ring image was lost, the uncertainty in measurement result will ascend in case of some what asymmetry. A compensation method based on neural network is presented. Then the basic performance parameters of the mockup are given.(3) A detailed analysis of uncertainty limit in RST is presented. Speckle is also the fundamental uncertainty factor in this kind of sensors. The analytic expression of centroid uncertainty limit is derived from the statistic of laser speckle, and it is shown that the uncertainty limit in RST, and also in anamorphic triangulation, is dependent on the solid angle subtended by entrance pupil as seen from the illuminated laser spot, the triangulation angle, as well as the laser wavelength. The influence of aberration of optical system and surface roughness are also discussed.(4) A method of getting the 3D edge characters based on the integrated vision system is studied. When using RST to acquire 3D information, a 2- dimensional scan is always needed. A novel symmetric multi-baseline matching based on the scanning movement is used to reduce matching error, and a subpixel matching based on quadratic interpolating is used to get a more smooth 3D edge. The matching is only done on character edges in images, so it is much more fast and robust.(5) A method of fast measurement path planning of RST guided by the 3D edge characters is studied. Because the RST sensor always measures in the direction vertical to the scanning plane, the 3D edge characters acquired by vision system can be projected to the scan plane and make a path planning of the RST sensor. Two planning methods are presented, one is driven by the 2-dimesional edges, and the other is driven by 3-dimentional surface patches rounded by edges. Through path planning, measurement points are much less than before, and natural quadratic surface patches can be analytically expressed.更多还原