Robust localization with wearable sensors

Thesis

Measuring physical movements of humans and understanding human behaviour is useful in a variety of areas and disciplines. Human inertial tracking is a method that can be leveraged for monitoring complex actions that emerge from interactions between human actors and their environment. An accurate estimation of motion trajectories can support new approaches to pedestrian navigation, emergency rescue, athlete management, and medicine. However, tracking with wearable inertial sensors has several problems that need to be overcome, such as the low accuracy of consumer-grade inertial measurement units (IMUs), the error accumulation problem in long-term tracking, and the artefacts generated by movements that are less common. This thesis focusses on measuring human movements with wearable head-mounted sensors to accurately estimate the physical location of a person over time. The research consisted of (i) providing an overview of the current state of research for inertial tracking with wearable sensors, (ii) investigating the performance of new tracking algorithms that combine sensor fusion and data-driven machine learning, (iii) eliminating the effect of random head motion during tracking, (iv) creating robust long-term tracking systems with a Bayesian neural network and sequential Monte Carlo method, and (v) verifying that the system can be applied with changing modes of behaviour, defined as natural transitions from walking to running and vice versa. This research introduces a new system for inertial tracking with head-mounted sensors (which can be placed in, e.g. helmets, caps, or glasses). This technology can be used for long-term positional tracking to explore complex behaviours.

Authors: Hou, X

• DOI: 10.5287/ora-jb1rpx1yy
• Type of award: DPhil
• Level of award: Doctoral
• Awarding institution: University of Oxford
• Language: English