Increasing Accuracy in Train Localization Exploiting Track-Geometry Constraints

Abstract

Train-borne localization systems as a key component of future signalling systems are expected to offer huge economic and operational advances for the railway transportation sector. However, the reliable provision of a track-selective and constantly available location information is still unsolved and prevents the introduction of such systems so far. A contribution to overcome this issue is presented here. We show a recursive multistage filtering approach with an increased cross-track positioning accuracy, which is decisive to ensure track-selectivity. This is achieved by exploiting track-geometry constraints known in advance, as there are strict rules for the construction of railway tracks. Additionally, compact geometric track-maps can be extracted during the filtering process which are beneficial for existing train localization approaches. The filter was derived applying approximate Bayesian inference. The geometry constraints are directly incorporated in the filter design, utilizing an interacting multiple model (IMM) filter and extended Kalman filters (EKF). Throughout simulations the performance of the filter is analyzed and discussed thereafter.