Research On Defect Detection Technology Of Three-dimensional Rail Surface Based On 2D/3D Composite Machine Vision

As a main part of modern comprehensive transportation system,railway transportation plays an important role in national economy.As important parts of railway tracks,rails with surface defects are prone to stress concentration under high speed condition of trains,which can even lead to fractured rails and cause great damage to the safety of life and property of people.Therefore,it is an important research topic for scholars and enterprises to detect defects quickly,accurately and efficiently in the production process of rails.There are many problems in defects detecting technology,such as poor accuracy,low efficiency,bad reliability,limited defects types,missing inspection,excessive detection,incomplete defect description,etc.So,it is difficult to apply in the defects detection fields of three-dimensional,complex and curved rails surfaces.Focusing on above scientific problems,the paper puts forward the defects detecting technology of three-dimensional rails based on the 2D/3D composite machine vision.Through the intensive study of defect characteristics,optical imaging,image processing,feature extraction and pattern recognition,the problems of key technologies are solved,such as the development of defect detection platform,the collection of high-quality rail surface image,the image processing under strong noise,the extraction of 2D/3D defect features and the high recognition rate of three-dimensional defects.Finally,the automated,non-contact,rapid,high-precision and comprehensive detection of defects on rail surfaces is realized.The main content and results are as follows:(1)The development of image acquisition system for curved rails and the establishment of evaluation system based on surface features and optical imaging models.On the basis of studying the characteristics of rail profiles,surface properties and optical imaging models,an image acquisition system for three-dimensional,complex and curved rails is developed.Through the design of six sets of binocular vision units and cross-sectional LED light sources,the synchronous collection and uniform illumination of surface images are realized.The comprehensive evaluation system of image acquisition system is also established according to the integrity,accuracy and quality of defect images.It can also ensure complete and high-quality image acquisition of surface defects;(2)Efficient two-dimensional features extraction of rail images under strong noise based on improved Sobel algorithm.Through the characteristics analysis of surface defects,a two-dimensional feature extraction scheme is developed.The wavelet transformation is used to smooth the noise produced in the process of image acquisition and signal transmission due to uneven exposure,fuzzy focus of optical systems and mechanical dithering.The suspicious defect area is determined by the image initial inspection,which is based on the local threshold change.Besides,an improved Sobel algorithm with multi-direction operator templates is proposed to accurately segment the defect area,which can also ensure accurate and rapid two-dimensional feature extraction of defect images;(3)Three-dimensional features extraction of dynamic feature point clouds based on SGA-FI-TM method.Two-dimensional features are short of a comprehensive description of the depth information for defect area.On the basis of studying binocular camera images and line-scanning camera calibration,it can match the defect area through three-dimensional spatial coordinate and parallax principle.As a result,feature-point-clouds are obtained in the defect area.Three-dimensional extraction scheme of dynamic feature-point-clouds based on SGA-FI-TM method is also put forward to reconstruct high-precision 3D defect model and extract six-dimensional features with depth information.Finally,effective acquisition of depth information and comprehensive characterization of surface defects on rails are realized.(4)The adaptive fusion of 2D/3D feature information and the design of efficient defect classification system based on SVM algorithm.Using the adaptive weighting fusion technology,2D/3D feature information of rail defects is integrated as inputting parameters of the classification system.Meanwhile,rapid and accurate classification of complex and highly nonlinear surface defects are realized by SVM algorithm after contrasting the classification efficiency and accuracy of different pattern recognition methods.(5)Preparation of detection platform for rail defects and rationality verification of research methods and theory in this paper.By simulating the industrial site environment,the detection platform for rail defects is designed.According to the comprehensive evaluation and results analysis in testing processes,the rationality and superiority of defect detection on rail surfaces are verified based on the 2D/3D composite machine vision.