EKF-based state estimation for train localization

Determination of longitudinal acceleration of a land-vehicle regardless its inclination is a common problem for systems of localization. This paper addresses a solution for railway applications by combining a low-cost MEMS IMU (Inertial Measurement Unit) equipped with a 3-axis accelerometer and a 3-axis gyrometer and an algorithm for data fusion. In particular, the impact of adding attitude and velocity observations into a Kalman filter is studied. Compared to conventional methods that use regular Kalman filter with external aiding sensors such as GPS or tachometers, the proposed approach uses an Extended Kalman Filter which exploits an augmented state vector. A velocity estimation obtained by a method observing the spectral analysis of the vertical accelerometer and the attitude estimation obtained by a complementary filter compose the observation vector with the accelerometer and the gyrometer data. At last, experimental results performed on an urban train are presented.