REAL TIME CLASSIFICATION OF RAIL DEFECTS
Massimiliano Nitti, Clelia Mandriota, Ettore Stella, A. Distante · Annual Conference on Computers · 2002
Maintenance of railway infrastructure can be accomplished using a number of different systems based on x-ray, ultrasound, or with image processing (IP), as has been used in this work. However, the IP method has some limitations such as rail surface localization and the algorithms used to detect the defect. Also, it is important that the light and camera are positioned to normalize the rail image. The camera and light were positioned as here to obtain a non-saturation rail image with the rail surface that is a part of the railway. Furthermore, the image contains the sleepers and all attached part of rail to the sleepers. Developing a system based on IP requires the solution of different tasks such as: railway tracking to extract the rolling plane surface and detection of defects. This paper describes the algorithmic solutions used for each of the above tasks, and the development of a prototype used to test the methods listed above.