Michael Lorenz

• With the rise of modern microelectromechanical (MEMS) systems - particularly MEMS-based inertial measurement units (IMUs) - inertial human motion capture has attracted increasing attention across various fields in recent years. With such systems, human motion capture tasks in sports, rehabilitation, or construction work have become feasible and accessible, but significant challenges remain. This dissertation presents analyses and advanced methods addressing two persistent challenges in inertial human motion tracking: disturbed magnetic fields and loose inertial sensor attachments. A disturbed local magnetic field leads to incorrect heading direction estimates. In indoor environments especially, magnetometer measurements can be unreliable due to disturbances such as ferromagnetic materials or electrical devices. The first contribution is an analysis of different orientation parametrizations in an optimization-based Bayesian smoothing setting, with only sparse magnetometer measurements provided. The second contribution is a method to completely avoid the use of magnetometers. We propose an approach to align sensors attached to the human body using only context information. The core idea is to estimate a common heading direction from a predefined motion sequence consisting of still standing and straight gait. Lastly, we explore a method to reduce the number of magnetometers needed to capture not only the joint orientations but also the absolute orientations of each sensor drift-free. Therefore, we extend a magnetometer-free Bayesian filtering solution for joint orientation estimation. In this manner, we are able to obtain drift-free absolute sensor and joint orientations with only one source of heading information in the kinematic chain instead of one per sensor. The second challenge covered in this dissertation is due to loose sensor attachments on the human body. Usually, inertial sensors are assumed to be tightly fixed to the body segments, which can be cumbersome in terms of setup time and ease of use. A setup where the sensors are directly integrated into one’s trousers or jacket might be a more desirable and comfortable solution. However, integrating the sensors into loose clothing usually results in additional clothing motion relative to the motion of the underlying segment that should be captured. To investigate the severity of such motion artifacts arising from the integration of sensors into a loose working jacket, we conducted a study of deviation angles between a setup with tightly attached sensors and loosely attached sensors. Based on this analysis, we propose a method called difference mappings, which aims to overcome the barrier of tightly attaching each inertial sensor to the human body. The proposed approach allows for reducing the joint angle errors due to clothing artifacts, thereby making an application with sensors integrated into loose clothing possible.