Multi-target Defect Detection of Railway Track Based on Image Processing

The tracks, sleepers and fasteners are easy to appear defects in the running process of track vehicles. Though the rail transit operation in China has developed rapidly, the development of track defect detection technology which main method is still manual inspection is relatively backward. Therefore, a method for automatic multi-target defect identification of railway track line using image processing has been proposed in this paper. Firstly, mean filtering and histogram equalization are used to denoise and enhance original images, respectively. After that, a Gamma correction based illumination compensation algorithm is proposed to realize automatic grayscale correction under the condition of uneven illumination. Secondly, an image recognition method based on scale-invariant features is adopted to extract the target features and to store them in the database. Then, the fasteners and track in the preprocessed images are located according to the statistical difference between grayscale and vertical projection. Finally, the gray-scale variance rate is utilized to extract the defects on the rail surface and the absence of rivets, and LBP (Local Binary Patterns) method is applied to extract the features of each part after segmentation, which are compared with the features collected in the database. The experimental results demonstrate that the accuracy of classification can reach 90%.