Detection Technology On Rail Surface Defects Based On Machine Vision

Because of the low detect ion efficiency,test results seriously affected by subject ivit y and other issues with current manual inspect ion of the rail surface defects detect ion, researches of the rail surface defect detect ion based on machine vision technology was carrieied out.In this paper,we realize automat ic ident ificat ion and classificat ion of rail surface defects by image processing, pattern recognit ion and other technologies.Combining machine vision theory with qualit y control requirements of the rail, we designed the overall solut ion of rail surface defect detection system based on machine vision and completed image acquisit ion using the hardware plat form built in the experimental environment; Besides, we sdut ied on the rail surface image processing algorithms, especially focusing on image preprocessing, defect first inspect ion and defect image segmentation. In the pre-processing stage, an adaptive method based on project ion integral to extract rail area was proposed according to the characterist ics of images.We uesd adapt ive median filtering to depressing the noise in rail surface images. In addit ion, the paper focused on the human visual attention mechanism and psoposed a defect segmentation method conbined context-aware visual inspect ion with iterat ive threshold to achieve defect segmentation; At last, we analysed the causes and features of the typical defects on rail surface, extracted defects' intensity, texture and geometry feature s and designed a mult i-class classifier based on support vector machine to recognize defects.By analyzing the results, the system can achieve the ident ification and classificat ion of rolling scars, rolling marks, inclusions and pseudo defects. The defects recognit ion accuracy rate can reach to 84%. Experiments show that the system has a certain reference value for pract ical use.