Research On The Detection Algorithm Of Railway Fastener Based On Gray Scale Invariance

Railway transport is an important part of Transportation Mode, so the maintenance of railway is very important to ensure its safety. As the connected component of rail and sleeper, railway fastener has an important influence on the safety of the railway. Image processing technology, as a kind of non-contact detection technology, gradually is used to railway fasteners state detection in recent years. In this article, the railway fastener defect detection is studied by using image processing technology. The main contents are as follows:First of all, Aiming at the poor adaptability and low accuracy of the existing railway fastener localization algorithm, this article puts forward a fastener positioning algorithm with gray scale invariance, which combines an edge feature enhanced method based on Rank transformation with a region growing algorithm.This positioning algorithm according the human eye visual acuity feature to set the size of the Rank transformation’s window, uses this window to traverse the fastener image for getting a Rank image, and then threshold the Rank image to enhance edge character. Later, this article uses the idea of the region growing method in the LSD combine with the priori knowledge to search the edges of the baffles in the edge enhanced image, then according the space position relations among the fastener, baffle and sleeper to position the fastener. Experiments show that this accurate algorithm has good adaptability and robustness.Secondly, In order to avoid the feature information redundancy problem of the original Pyramid Histogram of Oriented Gradients, this article proposes an improvement feature extraction algorithm based on it. According to the structure of the fastener image, this article proposes a new block partition algorithm on the basis of the original PHOG character, then extract the characteristics of the fastener by using this algorithm combine with the edge feature enhanced method which based on Rank transformation, and input them into the support vector machine (SVM) for training, in order to get the classifier model so as to realize the automatic identification of the fastener state detection. Experiments show that improvement feature can obtain better recognition effect and timeliness.