Low-cost MEMS-INS/GPS integration using nonlinear filtering approaches

Some important key issues in GNSS/INS integration mainly arise in the field of creating and developing low-cost, robust and at the same time highly accurate navigation systems, putting a focus of interest onto powerful sensor fusion algorithms. The so-called tightly-coupled integration is one of the most promising approaches to fuse the GNSS (global navigation satellite systems) data with INS (inertial navigation system) measurements. However, when modeling the underlying problem, the system process and observation models turn out to be nonlinear, and the GNSS stochastic measurement errors are often non-Gaussian distributed (e.g., due to multipath effects). Among other estimation approaches, the so-called particle filter (PF) as a nonlinear/non-Gaussian estimation method is especially theoretically attractive to be used in this field. However, its large computational burden usually limits its practical usage. In order to reduce the computational burden without degrading the system estimation accuracy, recently, an unscented particle filter (UPF) has been proposed, which combines the PF with the unscented Kalman filter (UKF). In this thesis, only one UKF is used in the algorithm, and the re-sampling step is not required anymore. Thus, the number of particles can be largely reduced, and the implementation of the PF on a hardware platform turns out to be feasible.