Estimating Magnetic Field at Joint Centers Reduces Kinematic Errors in Inertial Motion Capture.

Inertial measurement units (IMUs) are widely used to measure human motion, but accuracy remains inferior to gold-standard optical motion capture. Traditionally, researchers estimate joint angles from IMUs using global sensor fusion methods that assume the measured acceleration is gravity and the measured magnetic field is magnetic north. However, these assumptions are frequently violated, when linear accelerations are significant and magnetic fields are distorted. Recent magnetometer-free methods improve accuracy by replacing the gravity assumption with a more dynamically consistent assumption of a shared acceleration at the joint center, but these methods are prone to drift error. To improve accuracy, we developed the Magnetic Field at Inertial Joint Center (MAJIC) filter, which leverages both the common acceleration and magnetic field at a joint center. The magnetic field is adaptively included when necessary to reduce drift. We evaluated the MAJIC filter's estimated lower extremity joint kinematics against optical motion capture for 11 participants performing 20 minutes of ambulation tasks. The MAJIC filter produced joint angles with a median root mean squared error (RMSE) of 7.1°, compared to global sensor fusion methods (9.3°) and recent magnetometer-free methods (7.5°). The MAJIC filter also had a smaller range of RMSEs over all joints (5.5° to 11.4°) compared to global fusion methods (6.4° to 22.8°) and magnetometer-free methods (4.9° to 20.5°). We hope these improvements, and the open-source implementation of this filter, advance the measurement of human motion outside of the laboratory.