Localization and navigation technologies have become integral to modern society, playing crucial roles in daily life. They enable efficient and safe travel, allow emergency services to reach and assist individuals quickly, and are indispensable components of autonomous systems. Indoor localization technology, aimed at enabling precise location determination in indoor environments, has garnered significant research interest. One intriguing research direction is magnetic field-based localization technology, which exploits spatial variations in indoor magnetic fields to provide position information.

This thesis investigates how indoor magnetic fields can be used for localization and develops a magnetic field-aided localization system that does not rely on any preinstalled infrastructures, such as electric coils, or external localization information. To achieve this, a sensor platform consisting of a planar magnetometer array and an inertial measurement unit (IMU) was built. The array captures the spatial variations of the magnetic field, from which odometry information can be inferred. This odometry information is then used to aid an inertial navigation system (INS) constructed around the IMU on the array.

The thesis addresses three key challenges faced when realizing a magnetic field-based INS using the developed sensor platform. The first challenge is the calibration of the sensors to ensure their measurements are accurate enough for the localization system. The second challenge is to create a magnetic field model that can be used to realize a magnetic field-aided INS. The final challenge is to design a state estimation algorithm that provides consistent estimates so that the perceived uncertainties match the true estimation errors as closely as possible.

To address the first challenge, an easy-to-use and efficient calibration method is proposed to correct the misalignment of the IMU’s and magnetometer’s sensitivity axes, sensor biases, and scale factors. The second challenge is met by proposing a polynomial magnetic field model to construct a local small-scale magnetic field map and a tightly integrated magnetic field-aided INS. The proposed system was evaluated on simulation and real-world datasets, demonstrating a significant reduction in position drift compared to a stand-alone INS and showing performance comparable to state-of-the-art magnetic field odometry. Additionally, the system offers flexibility in sensor configurations, including sensor placement and the number of sensors involved. Finally, an observability-constrained magnetic field-aided INS is proposed to address the inconsistencies identified in the developed magnetic field-aided INS. This new system maintains the yaw angle unobservable, and demonstrates improved performance and consistency compared to the initial system.

The results show that the proposed magnetic field-aided INS can be realized by low-cost sensors and appropriate signal-processing algorithms. It could be integrated into magnetic field simultaneous localization and mapping (SLAM) systems to extend their exploration phase. Most importantly, it showcases the possibility of building self-contained, accurate, and consistent indoor localization systems with magnetic fields.