Study On Defect Recognition For Rail Surface Based On Machine Vision

High speed, high density and the heavy-haul of the railway traffic transportation is the trend. The safety of the railway plays an important part in transportation system. Railway surface detection can reduce the traffic accidents greatly. The digital image processing technology provides effective methods to deal with the railway surface defects. It’s a challenge to detect rail-steel’s surface defects in a vision system, because they appear randomly, have different sizes, distribute discretely, and under the influence of rust, noise points, and unequal illumination in imaging. According to the actual situation of rail defects, two detection methods were proposed.One method based on local contrast technique and the improved maximum entropy was proposed. Local contrast technique can enhance the rail surface image and reduce the jamming information at the same time. The improved maximum entropy which selects a threshold that maximizes the object entropy can get better segmentation results and reduce the computation complexity. Post-treatment to the segmented binary images by morphology operation was also implemented. This algorithm is fit in simulated system, which presents an accuracy of91.9%.There is a wide gap between the rail in real circumstance and simulated system, because the rail in use appears in high complexity, including more jamming information like featureless rust, unequal sunshine, noise points. One method based on gray level-gradient co-occurrence matrix and the maximum entropy was proposed. Gray level-gradient co-occurrence matrix was used for getting the mathematic metrics of the rail image. And the maximum entropy was used to get the threshold, which could be used to extract the inner border of the defects and can get good results. Post-treatment to the segmented binary images by morphology operation was also implemented. And the experiment results is satisfactory.