Marker-Assisted Platform Position Measurement Using Forward-View Train Images

This study proposes a marker-assisted method for measuring railway platform position using forward-view images captured from in-service trains. Conventional monocular-image-based approaches have limited applicability to precise infrastructure measurement because they suffer from depth-related uncertainty. To mitigate this limitation, the proposed method uses installed ground markers on the platform and sleepers, known marker dimensions, measured installation offsets, and track geometry information. The selected marker reference lines and points define a local transverse measurement plane under near-frontal imaging conditions. The method consists of YOLO-based marker detection, lens-distortion correction, DIC-based marker localization, local pixel-to-metric scale conversion, and vector-based geometric calculation. Field experiments were conducted on an operational regional railway line. When lens-distortion correction and the marker-center-based reference were used, platform gap estimation achieved an MAE of 4.6 mm, an RMSE of 5.3 mm, and a maximum absolute error of 8.8 mm. Platform height estimation improved after lens-distortion correction, with the MAE reduced from 14.2 mm to 9.0 mm, although the maximum absolute error remained 21.2 mm. These results suggest the feasibility of platform gap monitoring under the tested straight-track and near-frontal imaging conditions.