A precise train navigation system for Korean high-speed trains having a maximum speed of 400km/h or higher was developed in 2015. The navigation system employs multi-sensor fusion technique, and until now, the target performance of 1-meter instantaneous positioning accuracy in all environments including the Global Positioning System (GPS)-denied environments such as tunnels has been demonstrated using only simulations for various scenarios. In this paper, we demonstrate that the train navigation system achieves the target positioning accuracy in real train environments, which is the first train navigation system that the performance of sub-meter accuracy is demonstrated with real data in the literature. We also introduce additional filtering schemes applied to the various sensors of the train navigation system to enhance the sensor data corrupted by large unexpected noises often observed in the train environments. The train navigation system is properly modified for real train environments to employ the filtering schemes.