Maximum Likelihood Particle Filtering for the Fusion Direction of Arrival Beacons and IMU in Indoor Environments

Due to the wider availability of Global Navigation Satellite Systems (GNSS) signals in recent years, localization with low-cost devices is made possible. However, similar accurate and scalable positioning technologies are still in demand in GNSS-denied environments, such as indoor spaces. Some of the most ubiquitous indoor localization solutions are based on inertial measurement units (IMU) and beacons. In order to mitigate the errors affecting such solutions, a fusion technique can be used. Most state-of-theart fusions are based on the Kalman filter (KF). However, this filter assumes that the errors are Gaussian and is not suitable for indoor beacon-based localization with very few beacons. In order to improve the accuracy of the positioning, we propose a novel approach that relies on a small number of low-cost direction of arrival (DoA) beacons. The proposed method is implemented as a particle filter (PF). The computational cost of PF increases rapidly as the dimensionality of the solution space increases. In this research, this problem is addressed by reducing the navigation state from three to two (in planar space). This reduction is achieved by first maximizing the likelihood with respect to the agent’s (e.g., a robot, cart) heading and thus eliminating this parameter from the particle sampling space. The proposed solution can achieve up to centimetre-level accuracy.