Motion Analysis in Alpine Skiing: Sensor Placement and Orientation-Invariant Sensing

In alpine skiing, accurate and real-time estimation of body pose and inclinations due to turning is critical as it demonstrates the skier's turning behavior and abilities. Although inertial measurement units (IMUs) ease measuring kinematics in extreme conditions and provide such indications of skiers' behavior, they often suffer from sensor placement and orientation variability. This study explains the impact of sensor placement and orientation on the captured signals and proposes a preprocessing algorithm that can rotate raw signals from various locations and orientations similar to those near the Center of Mass (CoM). The preprocessing algorithm involves a sensor fusion approach using a quaternion-based complementary filter (CF) to rotate raw signals and extract turning motions via the global wavelet spectrum. Our experiment, validated on data collected from 14 sensors including two smartphones placed on different body parts during skiing sessions, demonstrates that the preprocessing algorithm can effectively reconstruct side motions, represent skiing turns, and detect turns independent of sensor placement and orientation. In field experiments with six skiers, the suggested preprocessing algorithm consistently detected skiing turns with an overall RMSE of 0.77 and MAE of 0.50 on all of the sensors relative to a reference sensor.