Efficient, simultaneous vision ray calibration and system orientation for high precision geometric-optical 3D-measurement systems

成像光学计量学允许高密度、高精度、快速和非接触地采集3D数据以进行质量控制。所使用的大多数方法的特征是光线沿直线传播，可以概括为术语几何光学测量技术。由于测量系统图像中的摄像头捕获的图像是二维的，因此不可能直接将点对点映射到3D坐标；相反，从每个像素发出半条线，即所谓的视觉或视线。通常，这些分配是通过基于所谓的传统针孔相机模型的相机校准来确定的。假设来自所有像素的视线穿过一个公共点，即所谓的针孔孔径。然而，真实的透镜系统不能用这种模型来充分描述。最近的一种方法分别描述了每个像素的视线。因此，校准为每个像素提供了单独的3D半线，这提供了可能的最高精度，但计算非常复杂。此外，参考记录的最小可能数量是未知的，理论预测的可能性被视觉射线校准技术的准无模型方法所限制。尽管如此，最常见的计量学相机校准方法是基于扩展的针孔相机模型。对于低不确定度几何光学测量仪器，这种方法是不能令人满意的，因为某些失真和透镜误差不能以这种方式捕获和校正。本项目的目的是通过开发新的、数值高效且几何最优的校准算法和策略，实现同时实现视光摄像机定标和多摄像机测量系统定向的新方法，从而大大简化基于条纹投影的多摄像机三维坐标测量系统的定标。本项目分为三个子目标：1.建立完整的多摄像机瞄准线标定仿真模型，考察模型各组成部分，验证所采用的数值优化方法，探索多摄像机标定方法。2.推导摄像机视线校准的规范，该规范描述了维持用当前使用的方法实现的不确定性所必需的最小参考显示位置集。测量时间将从几个小时减少到最多30分钟。3.开发一种用于多摄像机设置的校准方法，该方法通过同时校准和定向显著减少记录步骤的数量。在这里，对于双摄像头系统，测量工作量应减少到45分钟，而对于有四个摄像头的系统，测量工作量应减少到60分钟。

Imaging optical metrology allows high-density, highly accurate, fast and non-contact acquisition of 3D data for quality control. The majority of the methods used are characterised by light rays propagating in straight lines and can be summarised under the term geometric-optical measurement technique. Since the image captured by the camera in the measuring system image is two-dimensional, no direct point-to-point mapping to 3D coordinates is possible; instead, a half-line, the so-called vision or sight ray, emanates from every pixel. Usually these assignments are determined by a camera calibration based on the so-called conventional pinhole camera model. It is assumed that the sight rays from all pixels run through a common point, the so-called pinhole aperture. Real lens systems, however, cannot be adequately characterised by this model. A more recent approach describes the sight rays for each pixel individually. Thus the calibration provides an individual 3D-half-line for each pixel, which offers the highest possible accuracy but is computationally very complex. Also, the smallest possible number of reference recordings is not known, and the possibility for theoretical predictions is curtailed by the quasi-model-free approach of the vision-ray calibration technique. Still, the most common method of camera calibration for metrological purposes is based on an extended pinhole camera model. For low-uncertainty geometric-optical measuring instruments, this approach is unsatisfactory because certain distortion and lens errors cannot be captured and corrected in this way.The aim of the proposed project is the realisation of a new method for the simultaneous implementation of vision-ray camera calibration and multi-camera measurement system orientation by developing new, numerically efficient and geometrically optimal calibration algorithms and strategies that will greatly simplify the calibration of multi-camera 3D coordinate measurement systems based on fringe projection. The project is divided into three sub-objectives: 1. Creation of a complete multi-camera simulation model of sight-ray calibration to investigate various model components, to test numerical optimization methods used and to explore multi-camera calibration methods. 2. Derivation of a specification for the camera line of sight calibration describing the minimal set of reference display positions that are necessary to maintain the uncertainty achieved with currently used methods. The measurement time shal be reduced from several hours to a maximum of 30 min. 3. Development of a calibration method for multi-camera set-ups that significantly reduces the number of recording steps by simultaneous calibration and orientation. Here the measurement effort shall be reduced to 45 min for a two-camera system and to 60 min in a system with four cameras.