Research Of The Detection Algorithm On Railway Status Based On Image

With the development of the railway, the traditional inspection method is difficult to meet the requirement of modern railway maintenance. The automatic detection of railway state is the problem that must be faced. Some countries have developed several rail automation detection methods, the methods based on machine vision technology are the mainstream method. This automation detection technology can bring tremendous social and economic benefits due to its high efficiency, low cost and excellent adaptability.The objects of this paper are rail surface and rail fasteners. Machine vision technology is used to distinguish its current state. The main work is as follows:To avoid the deficiency of existing rail surface defect detection algorithm, this paper presents a new algorithm to solve this problem based on prior knowledge. This algorithm gets the precise area of the rail surface through the pre-positioning. Considering the image blur and uneven illumination, first it reconstructs the rail surface using improved robust locally weighted smoothing method and creates a defect threshold value table according to the statistical analysis. Then it can get the mutational image by contrasting the reconstructed rail image with the original image. So the rail defect region can be figured out by comparing the mutaional image with the threshold value table. The experimental result shows the algorithm has strong robustness and accuracy.By comparing different fastener state detection algorithms, this paper uses pattern recognition algorithm combined with histograms of oriented gradients to detecting the fastener images captured by our system. Considering the high dimension of original features and low efficiency, this paper improves the original feature and reduces its dimension by using kernel principal component analysis. Support vector machine is used to detecting the state of the fastener. In order to improve the recognition accuracy, fastener images are separated into several types and for each type the SVM is trained respectively, then every classifier is tested and verified.Finally the training sample library and the test sample library are established. This paper achieves the entire detection system on VS platform. The experimental result shows that the fastener state recognition algorithm based on image processing technology can meet the requirement of practical applications.