Drift Reduction for Inertial Sensor Based Orientation and Position Estimation in the Presence of High Dynamic Variability During Competitive Skiing and Daily-Life Walking

Nowadays inertial sensors are extensively used for gait analysis. They can be used to perform temporal event detection (i.e. step detection) and to estimate the orientation of the feet and other body segments to determine walking speed and distance. Usually, orientation is estimated from integration of the measured angular velocity. Prior to integration of measured acceleration to obtain speed, the gravity component has to be estimated and removed. During each integration small measurement errors accumulate and result in so-called drift. Since the first uses of inertial sensors for gait analysis methods have been presented to model, estimate and remove the drift. The proposed methods worked well for relatively slow movements and movements taking place in the sagittal plane. Many methods also relied on periodically occurring static phases such as the stance phase during walking to correct the drift. Inertial sensors could also be used to track higher dynamic movements, for example in sports. Potential applications focus on two aspects: performance analysis and injury prevention. To better explain and predict performance, in-field measurements to assess the coordination, kinematics, and dynamics are key. While traditional movement analysis (e.g. video analysis) can answer most of the questions related to both performance and injury, they are cumbersome and complex to use in-field. Inertial sensors, however, are perfectly suited since they allow to measure the movement in any environment and are not restricted to certain capture volumes. Nevertheless, most sports have very high movement dynamics (e.g. fast direction changes, high speeds) and are therefore challenging for computing reliable estimates of orientation, speed and position. The inertial measurements are compromised by noise and movements oftentimes don't provide static or slow phases used in gait analysis for drift correction. Therefore, the present thesis aimed to propose and validate new methods to model, estimate and remove drift in sports and for movements taking place outdoors in uncontrolled environments. Three different strategies were proposed to measure the movement of classical cross-country skiing and ski mountaineering, alpine ski racing, and outdoor walking over several kilometres. For each activity specific biomechanical constraints and movement dynamics were exploited. The proposed methods rely only on inertial sensors and magnetometers and are able to provide orientation, speed, and position information with an accuracy and precision close to existing gold standards. The most complete system was designed in alpine ski racing, probably one of the most challenging sports for movement analysis. Extreme vibrations, high speeds of over 120 km/h and a timing resolution below 0.01 seconds require maximum accuracy and precision. The athlete's posture and the kinematics of his centre of mass both in a relative athlete-centred frame and in a global Earth-fixed frame could be obtained with high accuracy and precision. Where 3D video analysis requires a very complex experimental setup and takes several hours of post processing to analyse a single turn of a skier, the proposed system allows to measure multiple athletes and complete runs within minutes. Thus, new experimental designs to assess performance and injury risk in alpine ski racing became feasible, greatly helping to gain further knowledge about this highly complex and risky sport.