IMU and magnetometer modeling for smartphone-based PDR

In this paper, IMU and magnetometer models that enable pedestrian dead-reckoning without step detection or zero velocity updates on hand-held devices such as smartphones are proposed. The models are suitable for usage with any standard Bayesian filtering or smoothing technique and thus, do not require customized estimators. The method is evaluated in a real scenario using a dataset of approximately three minutes with a trajectory length of 302m. It is found that the overall shape of the estimated trajectory matches well with the GPS reference trajectory with an estimated trajectory length of 293m. Some problems in the heading estimation are observed.