Object Detection and Binocular Depth Estimation in Railway Localization Applications

Résumé fourni par la source

Railway train operation status autonomous perception has demonstrated advanced ability of next-generation train control systems, for example perception accuracy improvement in satellite navigation-denied environments. This train borne centric status perception approach could reduce trackside equipment, thereby lower construction and maintenance costs while enhancing the autonomy and reliability of train operations. Relying solely on satellite-based positioning for autonomous train localization is highly susceptible to environmental influences. In contrast, computer vision-based positioning methods provide greater flexibility and are less affected by external conditions. This study leverages deep learning models to investigate the identification of key features of trackside POI (Point of Interest) during train operations, with recognition accuracy improved through color thresholding techniques. In distance perception tests, we compare conventional laser ranging methods, the Semi-Global Block Matching algorithm for binocular ranging, and deep learning-based estimation approaches. The results demonstrate that trains can effectively estimate distances to trackside markers without relying on GNSS or other localization systems The research offered insights for vision-based autonomous train perception and localization technologies.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.