Xueyin Liu,∗,∗∗ Chen Yan,∗ Yao Fu,∗∗∗ and Peng Chen∗
Railway track, wear inspection, point cloud registration, deep learning, sparse ICP (SICP) ∗ School of Mechanical Engineering, Southwest Jiaotong Uni- versity, Chengdu 610031, China; e-mail: [email protected]; [email protected] ∗∗ Sichuan Institute of Machinery Research & Design(Group) Co., Ltd., Chengdu 610063, China; e-mail:[email protected] ∗∗∗ Sichuan Center for Patent Examination Cooperation of the Patent Office of the State Intellectual Property Office, Chengdu 610213, China, e-mai
Rail wear is one of the key factors that determine the life of rails and directly affects the safety of railway transportation. Aiming at the rapid and accurate measurement of rail wear, an improved deep learning-guided sparse ICP (SICP) method is proposed for rail wear inspection in this paper. Firstly, the coarse point cloud registration based on deep learning is conducted as guidance to provide a better initial transformation matrix. In terms of accurate registration, a downsampling approach with voxel filtering and uniform sampling is proposed for the improvement of SICP. Finally, the rail wear amount is calculated according to the results of accurate registration. By analysing the influence of point cloud downsampling on the registration accuracy, time cost and rail head wear calculation accuracy quantitatively, it finds the proper numbers of points for downsampling to control the registration accuracy and time cost, and also reveals there is no effect on the wear calculation accuracy in such range of downsampling levels. Based on the comparison of noises, missing data, and different wear amounts in the rail point clouds, the robustness of SICP in the accurate registration for the rail wear calculation is verified. Simultaneously, the registration result of the switch rail demonstrated that the proposed method is also effective for point clouds with variable cross-section profiles. Compared with several classical and learning-based methods, it is proved that the proposed approach has higher precision and efficiency in rail wear measurement.
Important Links:
Go Back
