Probabilistic localization method for trains

Abstract

The localization of trains in a railway network is necessary for train control or applications such as autonomous train driving or collision avoidance systems. Train localization is safety critical and therefore the approach requires a robust, precise and track selective localization. Satellite navigation systems (GNSS) might be a candidate for this task, but measurement errors and the lack of availability in parts of the railway environment do not fulfill the demands for a safety critical system. Therefore, additional onboard sensors, such as an inertial measurement unit (IMU), odometer and railway feature classification sensors (e.g. camera) are proposed. In this paper we present a top-down train localization approach from theory. We analyze causal dependencies and derive a general Bayesian filter. Furthermore we present a generic algorithm based on particle filter in order to process the multi-sensor data, the train motion and a known track map. The particle filter estimates a topological position directly in the track map without using map matching techniques. First simulations with simplified particular state and measurement models show encouraging results in critical railway scenarios.