Research On Defects Inspection Technology For Rail Surface Based On Machine Vision

Security is the lifeline of rail tra nsport. Along with the opening line of highspeed, high dens ity and heavy-haul trains, the railway transportation safetyencounters a severe test. Preventio n and co mprehens ive ma nage ment is always theguide lines of the ra ilway sector. Under the influence of the traffic load, fatigue wearand the externa l environment, the surface of the rail produces a variety of de fects,which can ca use very b ig me nace for tra in sa fety. Because of co mple x terrain and along line of railwa y track, the actual effic iency of manua l inspection is very low andthere exist a lot of dangers and difficulties. So the research directions of the ra ilsurface detection s hould shift towards auto mated detection fro m a manua l inspection.In order to improve the speed and accuracy of defect detection, the paper des igns avisua l detection device, whic h can realize defects detectio n and recognitio n of the ra ilsurface.The paper introduces the research background and significance, expounds theresearch status of the track detection and the ma in contents and organizationa lstructure. Then it outlines the composition and applications of the machine vis io nsystems. The paper designed a set of rail surface defe cts automatic detectio n devicesand ana lyzed the light source, ca mera, optical lens, lighting solutions, mec hanica lconstructio n, acquisition and electrical contro l structure of severa l key techno logiesin detail.Under the influence of no ise and other external factors, the acquis ition of ra ilima ges can not be directly used for defect detection and recognition processing, so theacquis ition o f image need to be preprocessed operations. First, aiming at ima geenhance ment and filtering process, as we ll as for the lack of traditiona l filteringalgorithm and rail image characteristics, the paper proposed an adaptive filteringalgorithm to realize ima ge filtering processing. Further, in order to reduce the amountof subsequent image process ing operations, an image clipp ing a lgorithm wasproposed to achieve the posit io ning of the rail surface area. Fina lly, the research andanalys is of the common ima ge segme ntation algorithm provides a theoretica l basis forthe follow-up of the rail surface defect detection.The paper presented two types of defect detection process aga inst the rail surfaceima ge and screening the optima l algorithm by comparison study. One is based on a column-by-column scanning detectio n and the other is based on edge detection.Through the edge of the track with the characteristics of the defect information, itselects the appropriate feature vector to complete the learning of the BP ne ura lnetwork, and ultimate ly achie ve the class ification of cracks and scars. Theexperimenta l results show that the method can quick ly and accurately detect ra ilsurface defect, basica lly meet of technica l requireme nts of the system proposed. It hasa certain practical value.