Real-Time Vision-Based Wheel-Rail Interface Monitoring for Tramway Running Performance

Accurate monitoring of the wheel-rail interface is essential for safe tram operation and optimal vehicle performance. Conventional measurement techniques, ranging from strain gauges to gap sensors and earlier vision-based systems, can involve substantial hardware complexity, sensitivity to environmental disturbances, and onerous maintenance. To address these challenges, we propose a scalable camera-vision monitoring system tailored to tramways. A single high-resolution camera, combined with rigorous calibration and two stabilization modules (Wheel Contour Stabilizer and Rail Edge Stabilizer), converts pixel coordinates into reliable distance measurements while effectively suppressing noise. Wheel-rail boundaries are detected in real time by a YOLOv8-segmentation model trained on a customized dataset, ensuring robust performance across diverse lighting and weather conditions. Field trials on straight and curved tracks at speeds from 10 to 50 km/h demonstrate submillimeter accuracy (mean error $\approx $ 0.46 mm) and consistent robustness. The proposed approach facilitates straightforward maintenance and enables early detection of the risk of derailment. In addition, thanks to its camera vision foundation, it can be easily extended to additional rail applications. This work establishes a foundation for future research and rapid technology transfer in intelligent rail transportation.