【作者】 孙博；

【导师】 邾继贵；

【作者基本信息】 天津大学 ， 仪器科学与技术， 2018， 博士

【摘要】 轨道交通产业高速发展及制造业转型升级,对三维扫描测量需求不断提高,运动状态下高密度点云获取成为三维扫描测量研究中急需解决的重要难题。高速运动状态下测量对点云获取能力、测量速度、分辨率和连续性均提出了极高的要求。受测量原理限制或硬件性能制约,现有技术的发展水平与实际应用需求存在一定差距,目前仍缺少有效的测量方法。本文通过引入线阵图像传感和惯性传感技术,研究了一种运动状态下高密度点云获取新方法。原理层面,研究了线阵立体视觉测量方法,解决高速、高密度点云获取问题;将惯性传感技术引入到几何量精密测量研究中,研究了惯性位姿测量方法,解决高动态运动条件下高速、连续点云拼接问题。技术层面,研究了虚拟立体靶标标定方法,解决线阵相机高精度标定难题;研究了惯性累积误差补偿方法,控制点云拼接精度。论文的主要研究工作如下:1.以线阵相机成像模型基础,建立了线阵立体视觉空间几何和数学解析模型。通过引入匹配约束条件、利用静态随机二进制条纹投影技术、推导一维相似性度量函数及相关系数拟合公式并运用GPU并行运算实现了高速、高精度立体匹配。引入了线阵相机极线几何概念对匹配误差进行了定量分析,并分析了最佳交汇角度。实验验证了线阵立体视觉测量方法的可行性和精度。2.提出了基于虚拟立体靶标的线阵相机高精度标定方法。设计了小型平面靶标,通过将靶标摆放多个位置并结合面阵相机辅助测量构建了大型高精度虚拟立体靶标,获取了数目较多、分布范围较广的标定点。利用投影矩阵分解、非线性优化及全局优化算法进行了参数精确求解。实验证明了虚拟立体靶标标定方法比现有立体靶标标定方法具有更高的标定精度与可靠性。3.研究了基于惯性信息的点云拼接方法。概述了基于惯性信息的点云拼接方法的基本思想。研究了面向几何量精密测量、适用于点云拼接的惯性位姿测量原理。利用几何标定方法实现了IMU与视觉传感器相对位姿标定和初始对准。归纳了惯性位姿测量的误差来源和特性,通过建立姿态、速度和位置误差微分方程分析了惯性误差传播规律,为误差补偿提供理论依据。4.研究了点云拼接精度控制方法。阐述了基于惯性测量信息与外部参考信息数据融合的精度控制策略。研究了将卡尔曼滤波和卡尔曼平滑相结合的惯性误差最优估计与补偿方法。最后,实验验证了精度控制方法的有效性。更多还原

【Abstract】 The demand for three-dimensional(3D)scanning improves with the rapid development of the high-speed train and the evolution of the manufacturing industry.In-motion high-density point cloud acquisition has been an important problem in 3D scanning research.High-speed in-motion measurement makes new demands for point cloud acquisition capacity,speed,resolution and continuity.Existing techniques are restricted by their principle and hardware performance,and therefore cannot fulfill the emerging requirements.There is still no effective technique at present.This dissertation studies a novel in-motion high-density point cloud acquisition method by integrating line-scan imaging and inertial sensing.On the theoretical level,a stereo line-scan imaging approach is studied to solve the problem of high-speed and high-density point cloud acquisition.Moreover,inertial sensing technology is introduced into precise dimensional metrology.The fundamental principle of position and orientation estimation based on inertial sensing is studied for high-speed and continuous point cloud patching.On the technological level,a calibration method using a virtual 3D target is proposed for precise line-scan camera calibration.Besides,the error compensation technique for inertial sensing is investigated,to guarantee the accuracy of point cloud patching.The detailed contents are as follows.1.Based on the imaging model of line-scan camera,the geometry and analytical model of stereo line-scan imaging are established.High-speed and high-precision stereo matching is achieved by introducing matching constraints,projecting static random binary pattern,deriving one-dimensional similarity and correlation coefficient function,and using GPU parallel computation.The matching error is quantized based on the epipolar geometry of line-scan camera,and the optimal intersection angle is analyzed.The experiment verifies the feasibility and accuracy of the stereo line-scan imaging approach.2.A precise calibration method for line-scan cameras is proposed by using a virtual 3D target.A small planar target is designed.A large-scale precise virtual 3D target is built by placing the target into multiple positions,where an auxiliary matrix camera is used for monitoring the position and orientation of the target.A large number of calibration points with a wide distribution are obtained.The parameters are accurately determined by using projection matrix decomposition,nonlinear optimization and global optimization algorithm.The experiment results show that the proposed calibration method is more accurate and reliable than the existing calibration methods.3.A point cloud patching method based on inertial sensing is studied.The basic idea of point cloud patching is summarized.The basics and kinematics of inertial sensors for precise dimensional metrology and point cloud patching are studied.An initialization and alignment technique for inertial measurement unit(IMU)are proposed via mechanical calibration.The error sources and characteristics of inertial error are summarized.The propagation rule of inertial error is analyzed by establishing the differential equations of attitude,velocity and position error,which provides a theoretical basis for error compensation.4.An accuracy management technique for point cloud stitching is studied.The strategy based on fusing inertial measurement and external reference is described.The optimal estimation and compensation of inertial error are studied by combining Kalman filtering and Kalman smoothing.Finally,the experiment verifies the effectiveness of the accuracy management technique.更多还原