Research On Visual Inspection Algorithm For Railway Track Surface Defect

With the fast development of modern railway technology, the regular inspection of railway track is increasingly important. The traditional railway track inspection depends on manual inspection, which has been unable to adapt to the trend of high-speed and accurate automatic inspection. With rail as the object, a machine vision based automatic detection system for railway track surface defect is designed, which could implement the detection and identification of rail surface defect, showing a good prospects.Firstly, the background of the research on railway track defect inspection is illustrated, giving a striking demonstration of the necessary of the detection. By analyzing the current state and development trend of the research on railway track defect inspection, a machine vision based detection method is presented in this paper.Secondly, the causes and impacts of the existence of rail surface defects are analyzed in detail, and the requirements of function and performance of the inspection system are proposed. According to the general model of machine vision system and characteristics of rail surface defects, the overall structure of the machine vision based detection system for railway track surface defects is designed. A experimental simulation platform for rail surface defects detection is designed, to quickly verify the validation of the algorithm.Thirdly, the type and selecting method of light source, illumination and optical lens are analyzed. Besides, the type and parameter calculation of cameara are elaborated in detail. The optical imaging system, consisting of high-speed linear CCD camera and linear light source, is merged into the system to acquire rail images with consistent resolution at high speed.Then, according to the features of surface defect of railway track images, the detection algorithm for rail surface defect is researched, which could detect and mark the defect areas in the rail surface images.1) In the phase of image preprocessing, in order to reduce the computation time of the subsequent processing, the rail surface area is marked by using the method of horizontal projection.2) In the phase of rapid defect detection, because of the complex form of defect areas and the stable intensity of defect background, a combination algorithm including the method based on gray scale compensation and the method based on tophat operation, is proposed to search for the defect area quickly. By checking whether anomalous areas exist or not, it is decided that whether subsequence operations will be carried out for the binary image or not.3) In the phase of positioning defects accurately, related morphology algorithms are used to fill holes, and mark and extract defect areas from the binary image of railway track. Moreover, open operation is facilitated to bond a single defect area which is divided or spray-like. In the end, results of experiments show that the inspection algorithm could check out the majority of all defects quickly and accurately, showing the validity, rationality and accuracy to a certain extent.Finally, features of defect areas are extracted and the most effective features are selected. A trained BP neural network is used to classify the defects. The findings of exmperiments illustrate that the method of classification could distinguish scarring and corruged abrasion defect fastly and accurately, fulfilling the requirements of real-time and accuracy of the system.