点云频域配准的双目双线结构光列车轮对检测
发布时间：2018-07-31 09:17
【摘要】：针对单线结构光测量因范围有限以及遮挡等原因而无法检测整个车轮的轮缘及踏面的问题,提出了一种列车轮对双目双线结构光检测方法。该方法采用两组双线结构光传感器进行测量和数据采集。在两传感器产生的点云进行配准的过程中使用截面点云数据还原点云矩阵,然后对其进行傅里叶变换,在计算旋转角度时先进行极坐标转换,最后求取两矩阵的互功率谱,从而得到点云的旋转和平移矩阵。在此过程中考虑到点云噪声的存在,将辛格函数近似代替无噪声的互功率谱傅里叶反变换,从而确定在频域配准时噪声对配准参数的确定并无影响。接着采用相对值比较的算法和主成分分析拟合基准平面,对轮对直径方向进行标定。利用标定值对扫描数据进行处理,得出包括车轮直径、车轮滚动圆径向跳动等对尺寸参数。实验表明:点云频域配准的双目双线轮对检测技术对车轮直径的测量尺寸误差≤±0.05mm,车轮端跳的测量误差≤±0.06mm,车轮径跳的测量误差≤0.04mm。与标准轮对检测数据相比可知,该系统满足轮对检测精度的要求。
[Abstract]:In order to solve the problem that the single line structured light measurement can not detect the wheel flange and tread due to the limited range and occlusion, a new method of train wheelset binocular double line structured light detection is proposed. In this method, two groups of two-wire structured light sensors are used for measurement and data acquisition. In the registration process of point cloud generated by two sensors, point cloud data is used to restore point cloud matrix, and then Fourier transform is carried out. Polar coordinate transformation is carried out before calculating rotation angle. Finally, the cross-power spectrum of the two matrices is obtained. The rotation and translation matrices of the point cloud are obtained. In this process, the existence of point cloud noise is considered, and the Singer function is approximated to replace the noiseless cross-power spectrum Fourier inverse transform, so it is determined that the matching just-in-time noise in frequency domain has no effect on the determination of registration parameters. Then the relative value comparison algorithm and principal component analysis are used to fit the datum plane to calibrate the diameter direction of wheelset. The scanning data are processed by the calibration value, and the parameters including wheel diameter, wheel rolling circle radial runout and so on are obtained. The experimental results show that the measurement error of the binocular pair of wheels in point cloud frequency domain is 鈮，
本文编号：2155129
[Abstract]:In order to solve the problem that the single line structured light measurement can not detect the wheel flange and tread due to the limited range and occlusion, a new method of train wheelset binocular double line structured light detection is proposed. In this method, two groups of two-wire structured light sensors are used for measurement and data acquisition. In the registration process of point cloud generated by two sensors, point cloud data is used to restore point cloud matrix, and then Fourier transform is carried out. Polar coordinate transformation is carried out before calculating rotation angle. Finally, the cross-power spectrum of the two matrices is obtained. The rotation and translation matrices of the point cloud are obtained. In this process, the existence of point cloud noise is considered, and the Singer function is approximated to replace the noiseless cross-power spectrum Fourier inverse transform, so it is determined that the matching just-in-time noise in frequency domain has no effect on the determination of registration parameters. Then the relative value comparison algorithm and principal component analysis are used to fit the datum plane to calibrate the diameter direction of wheelset. The scanning data are processed by the calibration value, and the parameters including wheel diameter, wheel rolling circle radial runout and so on are obtained. The experimental results show that the measurement error of the binocular pair of wheels in point cloud frequency domain is 鈮，
本文编号：2155129
本文链接：https://www.wllwen.com/kejilunwen/zidonghuakongzhilunwen/2155129.html
