【摘要】 针对3D数据在离散与连续空间中多尺度处理导致极值点偏移量大，使得高速铁路轮对踏面擦伤面积分布映射结果与实际情况偏差较大的问题，提出基于3D扫描的高速铁路轮对踏面擦伤检测方法。利用高斯核函数构建高速铁路轮对踏面3D图像的尺度空间，通过DoG函数拟合局部关键点与尺度空间内的所有相邻像素点，校正极值点偏移量，输出3D图像的特征点云，避免了极值点偏移量的干扰。利用对称操作函数对特征点云进行高维映射后，根据多分辨率分组学习结果分割高速铁路轮对踏面3D数据中的擦伤语义。测试结果表明，所提方法在有效提取特征点云的基础上，能够实现对高速铁路轮对踏面擦伤的有效检测，擦伤面积分布与实际情况的偏差稳定在0.1 cm~2以内。更多还原

【Abstract】 Aiming at the problem of large offset of extreme points caused by multi-scale processing of 3D data in discrete and continuous spaces, which leads to significant deviation between the mapping results of high-speed railway wheelset tread scratch area distribution and the actual situation, a high-speed railway wheelset tread scratch detection method based on 3D scanning is proposed.Using Gaussian kernel function to construct the scale space of high-speed railway wheelset tread 3D image, local key points are fitted with all adjacent pixel points in the scale space through DoG function, correcting the extreme point offset and outputting the feature point clouds of the 3D image to avoid the interference of extreme point offset.After using symmetric operation functions to perform high-dimensional mapping on feature point clouds, the scratch semantics in the 3D data of high-speed railway wheelset treads are segmented based on multi-resolution grouping learning results.The test results show that the proposed method can effectively detect the tread scratches of high-speed railway wheelsets based on the effective extraction of feature point clouds, and the deviation between the scratch area distribution and the actual situation is stable within 0.1 cm~2.更多还原