Pose Adjustment Considering Linear Cross-Section Continuity of Line-Structured Light-Based 3D Tunnel Measurements
Pose Adjustment Considering Linear Cross-Section Continuity of Line-Structured Light-Based 3D Tunnel Measurements
复制标题
DOI：
10.1109/sii55687.2023.10039303
复制
发表时间：
2023-01
期刊：
影响因子：
--
通讯作者：
Takuya Igaue;Toko Hayamizu;Kenichi Yoshida;Satoshi Yamanaka;H. Asama;A. Yamashita
中科院分区：
文献类型：
--
作者：
Takuya Igaue;Toko Hayamizu;Kenichi Yoshida;Satoshi Yamanaka;H. Asama;A. Yamashita
本文提出了一种利用两架无人机进行线结构光三维测量时的位置估计误差减小方法。除了使用相机进行截面测量外，该方法还使用相机进行姿态估计。两个相机的姿态可以通过使用额外的相机执行2D-3D注册来估计。因此，通过交替移动两个摄像机，可以对隧道中的长段进行3D测量。在这种方法中，测量过程分为多个测量周期。这个过程会导致两个错误。首先，由于无人机不能完全停止在环境中，在横截面位置发生漂移。其次，由于截面位姿估计不准确，导致每个周期积分点云的旋转不准确。为了解决第一个错误，我们使用环境中的地标来稳定光线向量。为了解决第二个误差，我们基于在每个测量区域的关节处横截面在直线上移动的事实来补偿点云旋转。在实验中，我们对一个没有曲率的模型隧道进行了三维测量。该方法抑制了截面点云漂移。此外，在30 m的距离内，我们可以计算出隧道截面在轴向上的位置，误差为212 mm，其中误差最大。
In this paper, we propose a method to reduce the position estimation error of line-structured light-based 3D measurements using two unmanned aerial vehicles (UAVs). In addition to the camera used for cross-section measurements, the proposed method uses a camera for pose estimation. The pose of the two cameras can be estimated by performing a 2D-3D registration using an additional camera. Therefore, by alternatively moving both cameras, it is possible to perform 3D measurements of long sections in tunnels. In this method, the measurement procedure is divided into multiple measurement cycles. This process results in two errors. First, as the UAV cannot halt completely in the environment, a drift in the crosssection positions occurs. Second, the rotation of the integrated point cloud of each cycle is inaccurate owing to the pose estimation inaccuracy of the cross-section. To solve the first error, we stabilized the ray vectors using landmarks in the environment. To solve the second error, we compensated for the point cloud rotation based on the fact that the cross-section moves in a straight line at the joints of each measurement area. In the experiments, we performed 3D measurements of a model tunnel with no curvature. The proposed method suppressed the cross-section point cloud drift. In addition, we could calculate the position of the tunnel cross-section with an error of 212 mm in the axial direction, where the error was the largest, over a distance of 30 m.