[115] Computer vision for rail surface defect detection

P S Heyns, R Deetlefs, A J Oberholster, T R Botha, P S Els and D H Diamond

Department of Mechanical and Aeronautical Engineering, University of Pretoria, Pretoria, South Africa. 
Tel: 27 12 420 2432; Email: stephan.heyns@up.ac.za 

Computer vision is a very active area of research and is relevant to many industries. The progress in computer vision has assisted in the development of novel visual inspection techniques relevant to rail. Rail inspection using visual cameras has emerged as a popular technique with the advantage of high-speed application, low cost and appealing performance, and computer vision is regarded as one of the most attractive approaches for rail surface defect detection.
This paper reviews the nature of rail surface defects and considers the challenges associated with the identification of such defects. It briefly reviews previous work that has been conducted in the use of computer vision for railway condition monitoring. It then continues to introduce two novel techniques that have recently been developed for rail surface condition assessment. These include a digital image correlation-based approach, as well as a deep learning anomaly detection-based approach based on image data captured of a rail surface.
Keywords: anomaly detection, computer vision, deep learning, digital image correlation, rail condition monitoring, rail surface defect detection.