Innovative and efficient inertial navigation system for train localization in GNSS-denied environments

This paper proposes an innovative and efficient railway vehicle model to improve train localization in GNSS-denied environments. The inertial navigation system is mainly paired with a virtual sensor called transverse Zero velocity UPdaTe (ZUPT), which provides continuous zero-velocity information along the two transverse axes of the vehicle. This measure is often subject to bias, corrected in the case of a train by studying its dynamics and the impact of railway geometry. A second virtual sensor called automatic stop detection ZUPT provides zero-velocity information in the three dimensions when static phases of the vehicle are automatically detected. The transverse ZUPT parameters for bias corrections will be studied on thousands of hours of real data recorded along various French lines and environments (urban, regional, and high-speed trains) and with different trains over the past two years. The accuracy improvements of the positioning solution will be demonstrated on GNSS-denied urban data with horizontal errors divided by 4 from less than 20 m to less than 5 m in 99% of cases.