Research On Measuring Method Of Geometric Characteristic Parameters Of Rail Surface Defects Based On Machine Vision

China’s railway is developing rapidly,it has become the first choice for cargo transportation and passenger travel.However,the long-term operation of the train is more serious to the wear of the rails,and the quality of the rails directly affects the safety of the train.Therefore,regular inspection of the rails can timely find problems and reduce risks.At present,simple methods such as manual inspection are mostly used in China,which are easily affected by external factors and cause large detection errors.So,the need for efficient and accurate rail surface defect detection technology is more urgent.This paper proposes a method based on the combination of stereo vision and structured light scanning to measure the characteristic parameters of defects on rails with severely worn surfaces and comprehensively evaluate the defect status.The main research contents are as follows:Firstly,through the study of machine vision inspection at home and abroad,summarize a variety of methods for measuring the characteristic parameters of rail surface defects.Based on the profile characteristics of the tested rail,design a defect characteristic parameter measurement scheme which combining dual lasers and dual cameras.Analysis of measurement needs,the selection and construction of the experimental hardware platform was completed,and the software process was designed according to the measurement scheme.Secondly,extract the two-dimensional features from the defects on the rail surface image.The image is pre-processed first,and the bilateral filtering algorithm is selected to denoise the rail surface defects through subjective observation and data evaluation.Then compare the results of several segmentation algorithms,choose an adaptive threshold segmentation based on background difference for defect extraction,after that,calculate the six types of two-dimensional defect characteristic features for the acquired defects.Thirdly,extract the light curve after calibration.Considering the influence of various lighting factors,the collected RGB images of the light curve are converted to the HSV color space,and then extract the S channel images for segmentation,filtering and denoising.Then,use the improved Hessian matrix light strip center extraction algorithm to extract single-pixel rail profile curve.According to the method of obtaining the profile curve of the rail section without coordinate matching,the profile curves output by the two acquisition systems are fitted and compared with the non-defective profile curve,it can be seen that the height difference Δh of the defect site.Reconstruct the point cloud of each perspective,and stitch the rail point cloud according to the stitching method of the profile curve.Use the ICP algorithm to quickly register the reconstructed rail point cloud model with the rail point cloud in the defect-free area,this can get the defect area information.Finally,design the software according to the measurement system scheme,and a processing interface capable of human-computer interaction is displayed.To compare and analyze the defect feature results obtained by the measurement system with the results obtained by a handheld 3D scanner,it is proved that the stability and accuracy of the measurement system can meet the application requirements of non-contact rail surface defect characteristic feature measurement.Then verify the accuracy of the measurement system and analyze the experimental errors.