【作者】 晁志超；

【导师】 于起峰；

【作者基本信息】 国防科学技术大学 ， 光学工程， 2011， 博士

【摘要】 随着计算机、数字图像处理、自动控制等技术的发展,尤其是CCD和CMOS图像传感器性价比的不断提高,摄像测量以其客观、非接触、高精度等特点得到了迅速的发展和应用。然而,在利用摄像测量技术对大型结构体的位姿变形进行测量和监测时,会遇到两个难题:一方面,与激光陀螺、加速度计等高精度测量器件类似,摄像机也存在所谓的温度效应,而在长期测量的过程中不可避免的会遇到环境温度的变化,从而带来测量误差;另一方面,传统的摄像测量技术只能用于测量存在通视光路或处于同一视场内目标间的位姿关系,对科技或工程建设领域中大量存在的大型结构体上不通视或不处于同一视场的多个目标间的位姿关系测量则无能为力。本文对上述两个问题进行了研究,分别提出了用于减小摄像测量中温度效应的温度补偿方法和用于同时测量多个不通视目标间位姿关系的位姿传递像机网络摄像测量方法。在第二章中首先介绍了两种典型的摄像测量形式——双目位姿估计和单目位姿估计的原理以及它们在工程中的实际应用。由于单目位姿估计是像机链和像机网络中实现位姿关系传递的基础,因此着重介绍了目前常用的正交迭代位姿估计算法。此外,针对单目位姿估计在沿摄像机光轴方向上位移测量精度较低的缺陷,提出了一种利用高精度激光测距仪辅助的方法将两者的测量数据融合起来以提高系统的测量精度。在第三章中以经典的针孔成像模型为基础,以平面网格板为目标,对由温度变化引起的摄像机图像漂移与其内外参数变化之间的数学关系进行了研究。通过对模型中各个摄像机参数间的耦合进行仿真分析,推导出了不需要特定假设条件的线性图像漂移模型和适用于近距离应用下的简化图像漂移模型。分别设计了摄像机开启后自热阶段的图像漂移实验和通过开关空调改变环境温度的图像漂移实验,验证了图像漂移模型的正确性。实验中发现,摄像机开启后机身温度会迅速上升,引起摄像机内外参数产生相应的变化,大约1个小时后逐渐保持稳定,因此,在对摄像机进行高精度标定和测量时应当先开机预热1个小时,以减小摄像机自身发热带来的标定误差和测量误差。在第四章中对摄像测量中的温度补偿方法进行了研究。对基于系统辨识的线性动态系统建模方法进行了详细介绍,并利用该方法对温度变化环境下各个摄像机参数的温度变化模型进行了标定和验证实验,补偿结果显示,温度补偿能够显著减小摄像机的图像漂移量。为了验证温度补偿方法的可行性和有效性,进一步比较了单目摄像测量系统在温度补偿前后对平面目标的位姿测量数据,测量结果表明温度补偿方法能够显著减小由于环境温度变化引起的位姿测量误差。在第五章中对位姿传递像机网络摄像测量方法的基本原理及其表示方法进行了研究。首先介绍了像机链位姿传递摄像测量的原理和其在大型船体变形测量应用中的软硬件设计与实现,然后,提出了往返双像机链位姿传递摄像测量方法和基于平差的数据优化方法,最后,提出了位姿传递像机网络摄像测量方法,并讨论了图论算法在像机网络表示和数据优化中的应用。最后,在第六章中对位姿传递像机网络摄像测量中的两个关键技术——多目传递站位姿关系的最优标定和像机网络中的数据优化进行了研究。在分析多目传递站自身结构的基础上,提出了基于冗余标定的平差法优化标定方法。仿真结果表明,由于充分利用了多目传递站自身固有的各种约束关系,通过对原始的两两标定值进行平差法优化,可以显著提高多目传递站内各摄像机间相对位姿关系的标定精度。对于像机网络中的数据优化问题,提出了首先利用图论的方法得到像机网络中的有效约束条件,然后进一步提出了基于平差的数据优化方法和基于SPGD算法的数据优化方法。仿真结果表明,两种方法都能有效的利用像机网络中存在的固有约束,减小像机网络整体的测量误差,但是,平差法的复杂度会随着像机网络复杂度的增加而急剧增大,而基于SPGD算法的数据优化方法形式简单,复杂度基本保持不变,所以在大型结构体的位姿变形测量中具有良好的实际应用前景。更多还原

【Abstract】 Videometrics develops fast and is widely applied because of its impersonality, high precision and non-contact feature with the development of computer, image processing and automatic control technologies in recent years, especially for the image sensors such as CCD/CMOS becoming increasingly performance-cost. However, there are two difficulties encounterd in applying videometrics to measure the pose deformations of large structures. Firstly, video cameras have temperature effects as well as other high precision sensors such as the ring laser gyroscopes, the accelerometers, and so on, then thermal errors will be introduced when the videometrics apllying in a long period where the temperature of environment will change unavoidably. Secondly, traditional videometrics cannot be used to measure the pose deformation of objects in a large viewing field or non-intervisible objects in large structures, which are commonly used in technological and engineering construction today. To overcome the aforementioned difficulties, this thesis proposes a temperature compensation method to reduce the thermal errors in long time videometrics and develops the pose relay videometrics method using camera network to measure multiple non-intervisible objects.Firstly, principles and real applications of the monocular videometrics methods are introduced in chapter 1 as well as the binocular ones. Orthogonal iterative algorithm, which is one of the most popular monocular pose estimation algorithms, is presented in detail for the monocular pose estimation method is the base of pose relay in camera series and camera network. Furthermore, a new data fusion method is proposed to improve the precision of translation measurements along the optical axis of camera with a high precision laser rangefinding sensor.Secondly, mathematic relations between the image drift and the intrinsic/exterior parameters of the camera are studied targeted on the plane grid based on the classic pinhole camera model in chapter 3. The linear image drift model as well as the simplified one used in close range applications are farther proposed by analyzing and simulating the coupling between the varieties of camera parameters. Experiments of camera warm-up as well as experiments of environment temperature variations are implemented to validate the image drift models, in which the fact is found out that parameters of the camera rises immediately after camera startup for about 1 hour. So cameras had better to be warm-up for about 1 hour before calibrating and measuring in order to reduce the thermal errors introduced by the temperature rising of the camera.Then temperature compensation methods of videometrics are studied in chapter 4. Linear dynamic system modeling based system identification is particular introduced, with which the variety models of camera parameters are calibrated and verified under temperature variety environments. Experimental results show that image drift is reduced strongly under the temperature compensation method. Feasibility and validity of the method are further confirmed by comparing the pose estimations on plane object with/without temperature compensations.Principles of the pose relay videometrics using camera network are studied in chapter 5 as well as the representation methods. Software design and hardware realization of the pose relay videometrics using camera series as well as its application in the large ship deformation measurements are introduced firstly, then measurements of the pose relay videometrics using parallel camera series are studied and an adjustment based data optimization method is proposed to reduce the measurement errors. In the end, the pose relay videometrics using camera network is proposed, in which the graph theory is used in representing the camera network and optimizing the measurements.Lastly, the optimization calibration of multi-cameras relay station as well as the data optimization in the videometric using camera network are studied in chapter 6. An adjustment optimization method based redundancy calibrations is proposed by analyzing the structure of the multi-camera relay station. Simulation results show that the method could suppress noises effectively and improve the calibration precision because the full utilizing of constraint conditions in the relay station. Adjustment based method as well as SPGD (Stochastic Parallel Gradient Descent) based method are proposed in optimizating the measurements of pose relay videometrics system using camera network. Simulation results show that both methods could suppress noises and improve the measurement precision effectively because they can take full advantages of the constraint conditions intrinsic in the camera network, however, it is noticeable that complexity of the adjustment based algorithm will increase drastically as the complexity of the camera network increases, while complexity of the SPGD based method remains same, so the latter one is expected to find wide applications in the attitude angular deformation measurements of large structures.更多还原