A novel visual inspection system for rail surface spalling detection

Abstract

Railway transportation plays an important role in economic development since it can transport large volumes of passengers and cargo through great distances. Therefore, monitoring the condition of the railroad is essential to ensure train safety. With the development of computer vision, the railway non-destructive inspection systems have become possible. However, these methods have been still challenged by so many obstacles from ambient light on the rail surface and defects themselves. Defects appear on the rail surface are variety, in which spalling type is usually in heterogeneous shape and size. In this paper, a visual inspection system for rail surface spalling detection is proposed. The track image is first segmented by a novel rail track extractor. Then the rail surface spalling can be coarsely detected based on histogram curves in the longitudinal direction of the track image. Finally, a fusion technique is performed to eliminate all the false detected defects in the resulting image. Experimental results demonstrate that the proposed method achieves the precision and recall of 97.48% and 95.74%, respectively, and shows good robustness under nonuniform illumination and various rail surface conditions.