Bayesian GNSS/IMU tight integration for precise railway navigation on track map

Abstract

The localization of a train on a railway network with onboard sensors has been usually tackled by matching an estimated position fix on a track map. In this paper, we face the navigation problem directly in the topological domain of the map without computing an initial global position. We develop a tightly coupled scheme to fuse the raw measurement of Global Navigation Satellite System (GNSS), Inertial Measurement Unit (IMU) and the information of a digital track map. We model this problem using a Dynamic Bayesian Network (DBN) and we derive a particle filter to handle the discrete and nonlinear nature of the map in the estimation process and to detect the correct path after a train switch. Real measurements are finally used to test the method and analyze the role of raw GNSS measurements and satellite geometry in precision. Results suggest that reliable navigation can be achieved even when less than four satellites are in view.