Camera-based High-Speed Rolling Mark Detection

This paper introduces an automated system for high-speed rolling mark detection on railroad rails using computer vision. Rolling marks, engraved in the rail web, contain important information about the origin and quality of the rail. Captured using a color camera, these marks are identified using deep neural network object detection, specifically the YOLO architecture. The system can accurately interpret and correct detected characters, even if partially misidentified, by calculating a weighted Levenshtein distance. In this way, the system reliably identifies marks at high measurement vehicle speeds (up to 100 km/h), promising to enhance railroad infrastructure management.