基于随机纹理投射和单相机自由拍照的高质量三维外形测量关键技术研究

Considering the limitations of the state-of-the-art 3D measuring methods on large-scale, complex scenes, this project designs a new framework for 3D surface measuring based on the combination of random texture illumination and single camera free shooting, which exhibits a plenty of advantages, including simple hard structure, flexible measuring process, high accuracy result, automatic data registration, free expansion of measurement space. The key ingredients to be explored include: (1) Core problems of multi-view solving model are investigated; then a quantitative analysis of multiply images intensity constraint and an optimization method for sorting image sequence to establish stable 3D structure are proposed. As a result, more stable measurement field can be obtained and the registration and reconstruction errors can be dramatically reduced. (2) 3D reconstruction problem is systematically studied，and a new global linear cameras pose registration method based on images optimal-chosen mechanism is devised, an 3D reconstruction method with global optimum is proposed to guarantee the accuracy and stability of the result. (3) By using the complete structure information of images, we propose a coarse-to-fine matching algorithm, induced by the known multi-view geometry, between images under large deformation. Large deformation is parameterized and the matching problem is converted into a minimum deformation energy transfer. The result of this project will successfully provide a new theory and method on multi-view geometry, image matching and 3D reconstruction.

由于现有的视觉测量方法在大场景、复杂结构的三维外形测量上存在局限，本项目提出将随机纹理投射和单相机自由拍照相结合的高质量三维外形测量新方法，具有装置简单、过程柔性、结果精确、测量数据自动拼合、测量空间自由拓展等综合优点。研究内容包括：深入挖掘多视图几何求解模型中的核心问题，提出多图像间的约束强度定量分析和稳定支撑三维结构的图像优化排序，可显著提高测量场结构稳定性，极大的消除点云重建和拼合误差；系统研究三维重建过程，提出一种基于图像组合优选机制的全局线性化相机姿态定位技术，有效保证三维重建的精度和稳定性。充分利用多图像结构化信息，提出多视图几何结构诱导的大变形图像间稠密像点由粗到精匹配，对大变形进行参数化，将匹配求精看作是变形能量最小传递过程进行优化；本课题的研究成果将为多视图几何、图像匹配、三维重建等方向的提供新的理论和方法。

由于现有的视觉测量方法在大场景、复杂结构的点云测量上存在局限，本研究提出将随机纹理投射和单相机自由拍照相结合的高质量点云测量新方法，研究内容包括：1）设计出一套高质量的投影散斑生成算法，可对散斑颗粒的大小、密度、对比度和分布等参数进行调控。2）提出了一种带有旋转补偿的RFDIC算法。该方法利用图像间旋转变换信息，先对匹配图像进行旋转“校正”，使其满足DIC的相关性条件后，再利用DIC算法进行图像匹配。3）提出了一种无编码点的多视图几何求解方法，无需借助编码点信息而只需利用散斑纹理信息即可实现高精度的相机位姿求解的方法。4）以OpenMVG开源算法库为基础，完成多视图几何框架的建立，并实现了基于多视图的稠密点云的重建技术。通过对具体工业零件进行单相机拍照的点云重建综合实验，并将重建结果与专业扫描设备的测量数据进行对比，验证了本文的方法重建出的点云的精度水平。5）开发一套面向大型产品的工业摄影测量系统。以某型飞机实物作为试验对象，在10m×3m×4m的空间内，图像规模超过500张，测量误差低于0.085mm，测量结果与目前最新的商用工业摄影测量系统进行了对比，在精度上达到了相同水平，此方法在效率上更具优势，且具有更好的扩展性。