Train-borne Localization Exploiting Track-Geometry Constraints -- A Practical Evaluation

TL;DR

This paper evaluates a train-borne localization method that leverages track-geometry constraints, demonstrating practical accuracy improvements over standard approaches through real-world testing and data analysis.

Contribution

It presents a practical evaluation of a track-geometry constrained localization algorithm, advancing train-borne localization accuracy and assessing its real-world applicability.

Findings

Improved localization accuracy using track-geometry constraints

Successful validation with GNSS and IMU data from test drives

Effective estimation of geometric track-map during localization

Abstract

Today's railway signalling system heavily relies on trackside infrastructure such as axle counters and track balises. This system has proven itself to be reliable, however, it is not very efficient and, moreover, very costly. Thus, it is not suited to overcome the future challenges in railway transportation. For this reason, signalling systems based on train-borne sensors have gained interest recently. In this context, train-borne localization is one of the main research challenges. So far there is no sensor set-up which meets the demanding requirements for a localization system, both in the sense of accuracy as well as safety. To help overcome these issues in the near future we present our latest research results here. Earlier we published a localization algorithm which is characterized by an increased accuracy in cross-track direction compared to a standard Kalman filter (KF) approach, as has been shown in simulations [1]. To verify these results practically, we recorded data from a Global Navigation Satellite System (GNSS) and an inertial measurement unit (IMU) on a test drive. The localization accuracy is evaluated with the help of OpenStreetMap (OSM) data and site plans. Furthermore, we evaluate the quality of the estimated geometric track-map, which is additionally provided in the process of the localization algorithm [2]. We conclude with some remarks on the research challenges towards train-borne localization and suggest further steps to overcome them.