A 3-D Reconstruction Method for Large Freeform Surfaces Based on Mobile Robotic Measurement and Global Optimization
A 3-D Reconstruction Method for Large Freeform Surfaces Based on Mobile Robotic Measurement and Global Optimization
复制标题
基于移动机器人测量和全局优化的大自由曲面三维重建方法
DOI：
10.1109/tim.2022.3156205
复制
发表时间：
2022
影响因子：
5.6
通讯作者：
Zhouping Yin
中科院分区：
文献类型：
--
作者：
Jinshan Wang;Zeyu Gong;Bo Tao;Zhouping Yin
自动测量和三维重建是大型曲面零件机器人制造的关键。现有的方法在测量范围、自动化程度和重建精度等方面存在局限性。为实现大型自由曲面零件的高精度、高效率自动测量与重构，提出了一种基于移动的工业机器人的测量方法和一种基于多视点云配准的三维重构方法。首先，采用移动的测量机器人对大型自由曲面零件进行自动测量，获得局部测量点云。利用机器人的位姿信息和激光雷达的定位结果估计出初始的全局测量位姿，对局部多视点云进行粗拼接。其次，提出了一种基于点云碰撞检测的邻接判断方法，并利用该方法构造了局部测量位姿约束。最后，提出了一种基于改进姿态图优化的全局优化方法，实现了多视点云的全局测量姿态精确估计和高精度重建。对标准试验叶片和真实的涡轮机叶片进行了测量和重构实验。结果表明，该方法在4.5\times2.5\，\，\text {m}^{2}$的测量范围内，平均测量精度达到0.835 mm，在测量范围、测量精度和实用性方面均优于以往的方法。
Automatic measurement and 3-D reconstruction are critical to the robotic manufacturing of large surface parts. The existing approaches have limitations in terms of measurement range, automation degree, and reconstruction accuracy. To achieve the automatic measurement and reconstruction of large freeform surface parts with high accuracy and high efficiency, this article proposes a measurement approach based on a mobile industrial robot and a 3-D reconstruction method based on multi-viewpoint cloud registration. First, to perform the automatic measurement and obtain the locally measured point clouds of the large freeform surface part, a mobile measurement robot is employed. And the initial global measurement poses, estimated by the robot’s pose information and the light detection and rangings (LiDAR’s) location result, are used to coarsely stitch the local multi-viewpoint clouds. Second, an adjacency judgment approach based on point cloud collision detection is proposed, with which the local measurement pose constraints are constructed. Finally, a global optimization approach based on improved pose graph optimization is proposed to achieve accurate global measurement poses estimation and high-accuracy reconstruction of multi-viewpoint clouds. Experiments were carried out on measuring and reconstructing a standard test blade and a real wind turbine blade. The results demonstrate that the proposed approach reaches the mean accuracy of 0.835 mm for a measuring range of about $4.5\times2.5\,\,\text{m}^{2}$ , which outperforms the previous methods in terms of measuring range, accuracy, and practicality.