Indoor localisation is a growing field of interest in recent studies. While GPS (global positioning system) is a standard for outdoor localisation, no such solution exists for indoor applications. The literature provides several methods to obtain the location of indoor systems, often using optical sensors. A small number of recent studies use the indoor magnetic field for localisation. Ferromagnetic materials in the structure of buildings cause magnetic anomalies that are distinct enough to use for localisation. To use the magnetic field for localisation, a map has to be created.
In this thesis, a sensor setup different from other studies is used. This sensor setup consistsof a inertial HMTS (human motion tracking suit), containing seventeen IMUs (inertial measurement units) with magnetometers. This suit uses advanced techniques to obtain a better pose estimate than a single IMU can achieve. The combination of an inertial motion tracking suit with a magnetic localisation approach has not been studied before. Inspired by the state of the art approaches, a SLAM (simultatenous localistation and mapping) algorithm is proposed that is able to use information from the inertial HMTS. The algorithm consists of a reduced rank GP (Gaussian process) to create a map of the magnetic field. A RBPF (Rao-Blackwellized particle filter) is used to localise the HMTS. The algorithm allows for the use of multiple magnetometers to create the magnetic field map instead of a single one, which is a novelty.
The proposed method is tested with real-life data. Live odometry obtained from the HMTS can be post-processed for improved inertial odometry. Both the live and post-processed odometry data from the HMTS is used in the proposed algorithm and the results are compared. Additionally, the differences between a magnetic field map constructed with a single magnetometer and a map constructed with multiple magnetometers is investigated. The trajectory estimated by the RBPF is compared to a groundtruth, obtained by an optical tracking system. The RBPF shows higher performance for trajectories longer than 250 seconds compared to the inertial odometry. The use of multiple magnetometers does not improve the performance of the algorithm.