Algorithms for relative train localization with GNSS and track map: Evaluation and comparison

Abstract

Large safety distances between consecutive trains limit the flexibility and capacity of rail traffic. The introduction of automatic distance control between trains has the potential to reduce the safety distances. This requires an accurate and reliable distance estimation. In this paper, we therefore evaluate the performance of a relative localization system that tightly integrates the global satellite navigation system (GNSS) measurements and a map of the track network. The tight integration reduces the absolute train position to a 1-D value that describes the position on a track. The relative position is then calculated by subtracting two 1-D positions. In an empirical evaluation the tightly integrated system is compared to a loosely integrated system and a cooperative approach based solely on GNSS measurements. The cooperative approach uses GNSS pseudorange and range rate double differences, to determine the baseline between two antennas. For the loosely integrated system first the 3-D train position is estimated in an extended Kalman filter (EKF). In a second step, this position is matched to the map to obtain the position on the track. The relative position root-mean-square error (RMSE) of the different approaches is determined with a 167 km long data set, measured on a diesel train over the duration of 6 hours. The data set is divided into open sky, suburban and forest environments. For each of these environments, six different antenna distances are evaluated. The results show that the tightly and loosely integrated systems have a considerable smaller RMSE than the cooperative approach. The difference in the average performance of the two map based approaches is negligible. An advantage of the tight integration can be seen only under poor satellite visibility conditions that were encountered only sporadic during the measurements.