Reshearch On Image Detection System Of Metro Rail Missing Fasteners

Rail fasteners are important parts of rail line, which, once missing will cause a great threat to railway safety and may even lead to serious accidents. The detection of the railway fastener state in our country those days still depends on manual inspection, which is low efficiency, strong subjectivity and the high missing rate, because it relies on the worker’s technical proficiency. Currently, the rapid development of the railway, in particular high-speed rail, makes these problems become increasingly prominent. How to quickly and accurately detect missing fasteners is an urgent problem to be solved.In recent years, computer technology and image processing technology have developed rapidly, and the automatic detection of railway fasteners state based on computer vision has been the focus of current research both at home and abroad. The system is advantageous for it is non-contact, fast, highly accurate, and strongly adaptable, etc. According to present study on fastener state detection at home and abroad, this paper proposes an automatic detection system based on computer vision, focusing on designing the detection system scheme, building the hardware system, analyzing and designing the software system, and researching fasteners rapid detection algorithm, finally to achieve the automated online testing of missing railway fasteners. The specific work is as follows:The exact location of fastener from original image is an important prerequisite of the detection system. For the deficiencies of the existing location algorithm, the paper proposes a cross localization method to locate rail and fasteners area. First, The wavelet soft-threshold denoising method are used to reduce the interference effectively; secondly, a method for edge detection based on Canny operator was used; thirdly, making horizontal and vertical projection, then using regional scanning statistical method to locate the boundaries of the rails and sleepers, ultimately determining the fasteners area. The results show that the method is accuracy, robustness and stability.After edge detection of the fasteners image, firstly, deal the binary image with connection and morphological methods, then extract features, including:image area, image Euler, image entropy, the time domain statistical features of the vertical projection of the binary image(including:mean, standard deviation, mean square deviation, kurtosis, skewness), and the regional area characteristics of segmentation image. These features not only accurately reflect the fasteners states, but also greatly reduce the computation amount of matrix features calculation, and reduce the computation amount of the system; thereby the classifier training and recognition speed are improved.For normalized processing of the extracted features, this research applies the BP neural network and the fuzzy C-mean clustering to classify the fasteners. The two algorithms can achieve the fastener state classification. Recognition algorithm can be used according to the actual conditions.In the application of pictures collected in worksite, whose results show that:the features extraction algorithm of fasteners in this paper has strong stability, can avoid being influenced by illumination and rail tilt, can extract fasteners’region quickly and accurately; the image processing methods and classification algorithm can effectively identify fasteners sate, are of fast detection speed. It can replace the traditional manual inspection to a certain extent.