3D tensor-based point cloud and image fusion for robust detection and measurement of rail surface defects

Railway transportation safety relies on the accurate location, detection, and measurement of rail surface defects. However, the absence of image and point cloud fusion methods to address this challenge is a significant limitation. This paper introduces a 3D tensor-based point cloud and image fusion (T-PCIF) method, utilizing image and point cloud analysis for robust defect detection and 3D measurement. The defect region is extracted through the application of tensor robust principal component analysis (TRPCA) and eigenvalue decomposition on the constructed tensors, employing the KS partition. Experimental evaluations conducted under various parameters demonstrate that the T-PCIF method achieves an accuracy rate of 86.27% MDPA and 0.7018 MDIoU values.