Demonstration of optical metrology for rail vehicle and track condition monitoring

Accurate measurement of the lateral displacement between railway wheels and rails is critical for assessing track alignment and ensuring the safe operation of rail vehicles. This paper presents a novel, non-contact methodology that employs a single stereo camera to monitor the wheel-rail interface. Video frames extracted from the left camera are processed using Segment Anything Model 2 (SAM2), a zero-shot deep learning segmentation model, to isolate the wheel and rail regions without the need for prior training or annotation. The selected image coordinates are then mapped into three-dimensional space using the ZED SDK’s Neural Depth engine, enabling real-time estimation of lateral displacement. The system was tested on a diagnostic railway vehicle under controlled low-speed runs, demonstrating reliable segmentation performance and consistent displacement measurement. The proposed approach eliminates the need for manual labeling, training datasets, or physical markers, offering a fast and scalable solution for onboard track geometry monitoring.