Map Based Multisensor Railway Localization Enhanced by Raw GNSS Data

Kurzfassung

Precise train localization is essential for railway operation, train control, track management and collision avoidance. Therefore, the positioning of railway vehicles is a safety critical application. Nowadays, these applications rely on track-side equipment and suppose a high cost in development and maintenance. By contrast, train-side mounted sensors, such as Global Navigation Satellite Systems (GNSS) can achieve railway positioning at a low cost. In particular, the European satellite navigation system Galileo enables safety related applications by new Safety-of-Life (SoL) services. However, GNSS suffers from line-of-sight problem and its availability and accuracy are dependent of the number of visible satellites. Moreover, stand-alone GNSS can not provide the required precision in parallel track scenarios. Therefore, in order to achieve a reliable and robust localization for trains, additional train-side sensors and information of a railway map are combined with GNSS measurements. In this thesis, a probabilistic approach for train localization is proposed by using a Bayesian filter. In particular, a particle filter estimates the topological position of the train in the railway network from raw GNSS measurements (i.e., pseudoranges and Doppler shift frequency), Inertial Measurement Unit (IMU) data and the track map, resulting in a tightly coupled integration. The proposed approach is compared to a loosely coupled integration that uses the PVT solution of the GNSS receiver combined with IMU and the map. Finally, an evaluation of the tight and loose approach with data recorded in a railway environment is performed. Results are very encouraging and suggest high reliability of the proposed method even in low visibility situations.