Research On Image Processing Technology Of Automatic Rail Defect Detection System

Rail flaw detection and fastener inspections are important parts of the railway infrastructure detection. The automatic rail defect detection system based on image processing can be an efficient way to detect railway defects automatically and on real-time.The article focuses on image processing technology of the rail defect detection system, the main works are as follows:According to the objectives and requirements of the defect detection system, the main equipments of the image acquisition system were selected and the rail defect detection hardware part was composed. In addition, a software part for automatic rail defect detection system was analyzed and designed, including image acquisition and control system, image analyzing system and data management system.The extraction of the rail track and fasteners is an important prerequisite of defect detection. For the deficiencies of the existing location algorithm, an improved location algorithm based on regional highlights statistics was proposed. Using the same method the edge of the rail track and the edge of the sleeper were positioned. Thus the fasteners could be extracted using the prior knowledge of the fastener’s relative position to the rail-sleeper cross.Threshold segmentation, morphological methods were used to preprocess the extracted rail track, and the normal rail track could firstly be excluded here. For the case of defected rail track, the defect’s features such as centroid, area, perimeter, length-width ratio were calculated. Proper features were chosen, and a3-layer BP neural network was designed to classify these two defects as scars and cracks.PCA dimension reduction method was used to extract the fastener features, thus greatly reducing the computing amount of fastener present/absent identification. The training library and the test library of fastener pictures were established, and with the use of the nearest neighbor classifier the fastener present/absent pattern recognition was realized. The experiments show that the PCA method can achieve a high recognition rate even when the fasteners are of different size and with small amount of occlusion.Based on MATLAB tools, experiments have been done to verify the above algorithm. Experiments show that the rail and fasteners extraction algorithm has strong stability. The rail surface defect detection algorithm can exclude normal rails, BP neural network designed in this article can effectively identify the scars and cracks; PCA dimension reduction feature extraction method and the nearest neighbor classifier method is of a certain robustness and can quickly and efficiently identify the fastener present/absent pattern.