图像式路基监测系统中站内多相机间位姿标定方法

Calibration Method of Position-Pose Relation Between Cameras in Transfer Station of Image-Based Subgrade Monitoring System

相机间位姿关系是影响图像式无砟轨道路基沉降测量精度的主要因素。监测站中相机必需指向各自的监测靶面,因此相机间常存在无公共视场的情况。根据机器视觉中手眼标定问题与相机间位姿标定的等价性,提出了一种基于特征点的相机间位姿标定方法。通过在监测靶面上设置4个呈方形分布的特征点,使相机组执行2次运动,并在3个不同的位置拍摄靶面图像,通过P4P(perspective-four-point)算法求解相机与靶标的位姿关系,进而得到相机的移动轨迹;用矩阵重排的方法求得相机间位姿转换矩阵,并利用Levenberg-Marquardt算法进行非线性优化。仿真结果表明噪声方差在1pixel内时角度误差小于0.03°,平移误差小于0.3cm。通过仿真验证了本文方法的有效性与实用性,本文方法能够满足无砟轨道路基沉降测量的需求。

The position-pose relation between cameras is the main factor that affects the accuracy of an image-based ballastless track subgrade monitoring system.Each camera in the monitoring station must face the corresponding monitoring target surface,so there is often no public field of view between the cameras.According to the equivalence between the hand-eye calibration problem in robot vision and the calibration of position-pose relation between cameras,aposition-pose calibration method based on feature points is proposed.Four feature points with square distribution are set on the monitoring target surface.A set of cameras is moved twice in small step and takes the pictures of the target surface in three different positions.The P4P(perspective-four-point)algorithm is used to solve the position-pose relation between the camera and the target,and then obtains the movement trajectory of the camera.The matrix rearrangement method is used to obtain the pose conversion matrix between the cameras,and the Levenberg-Marquardt algorithm is used for nonlinear optimization.Simulation results indicate that the angle error is less than 0.03°and the translation error is less than 0.3 cm when the noise variance is less than 1 pixel.The effectiveness and practicability of the method in this paper are verified by simulation,and the method in this paper can meet the needs of the settlement measurement of the ballastless track subgrade.