An Approach to Magnetometer-free On-body Inertial Sensors Network Alignment Michael Lorenz ∗ Bertram Taetz ∗ Gabriele Bleser ∗ ∗ Technische Universi¨at Kaiserslautern, Kaiserslautern, Germany (e-mail: {surname}@cs.uni-kl.de). Abstract: To capture human motion with inertial sensors, they are attached as a network on different segments. Typically the measurements received from each sensor are fused to obtain its orientation. A challenging task is to align the orientation of each sensor w.r.t. to a single common coordinate frame. To fulfill this task typically the local magnetic field is measured to provide information about the heading direction. Since especially in indoor environments magnetic field disturbances can be present, this information is not a reliable source. To overcome this problem, we present a method that aligns an on-body inertial sensor network using gyroscopes and accelerometers only. The subject wearing the network had to fulfill a predefined procedure, consisting of standing still and walking straight. To extract the heading direction, we estimated the linear acceleration and angular velocity using a maximum-a-posteriori estimator. Performing a principal component analysis on the estimated states we computed two heading directions for each estimate. Instead of using them separately, we used a fusing approach that exploits symmetrical effects. We validated the approach on a lower body configuration using an optical motion capture system. The heading direction of sensors attached on a single leg could be aligned up to median maximal deviation of 2.6 degrees and on the complete lower body of 6.6 degrees. Especially deviations of the pelvis were higher, due to a lack of motion excitation. To be able to quantify the excitation needed, we proposed an indicator based on the ratio of the eigenvalues of the principal component analysis of the angular velocities. Keywords: Human body motion capture, inertial sensors, sensor network, sensor alignment, spatial synchronization, motion estimation, information and sensor fusion, parameter and state estimation. 1. INTRODUCTION The microelectromechanical systems (MEMS) technology allows to construct inertial measurement units (IMUs), also called inertial sensors, which are small, light-weighted and have a low power consumption. As Aminian and Najafi (2004) pointed out, this allows to attach them on the hu- man body and apply them in numerous applications such as sports, rehabilitation or daily life monitoring. Usually MEMS IMUs are also equipped with a magnetometer, which allows them to measure the local magnetic field. Combining several IMUs in a sensor network attached to the human body, they are used in order to solve different tasks, like human body motion tracking or human motion analysis, as for instance done by Teufl et al. (2019), von Marcard et al. (2017) and Prathivadi et al. (2014). Usually such tasks are solved by first estimating the orientation of the moving body parts w.r.t. to a common coordinate frame (L), see Fig. 1. In a subsequent step the orientations are used to resolve the original problem. Regarding gold standard systems for human motion capturing, meaning optical motion capturing systems based on reflective mark- ers, the primal task of obtaining the orientation w.r.t. to a reference coordinate frame is comparably simple. As shown in Guerra-Filho (2005), optical systems are able to capture the positions of markers w.r.t. to a local coordinate frame Z X Y L I 1 I 2 I 3 I 4 I 6 I 7 I 5 sagittal plane Fig. 1. The figure shows the experimental setup for the validation of our approach using seven IMUs and an optical motion capturing system. The common local coordinate frame (L) is used to orientate each IMU. Preprints of the 21st IFAC World Congress (Virtual) Berlin, Germany, July 12-17, 2020 Copyright lies with the authors 16203