Lower Limb Sensor-to-Segment Calibration for Joint Kinematics Analysis With Inertial Measurement Units: Is There an Ideal Method?

A prerequisite for obtaining human movement kinematics from inertial measurement units (IMUs) is to define the relative orientation between the IMU coordinate systems (CSs) and that of the underlying segment: a step called sensor-to-segment (S2S) calibration. Many S2S calibration methods have been proposed in the literature for the lower body segments. However, these methods were not compared in a single study and the methodology differences between the studies make an objective assessment of the proposed S2S methods impossible. The present study aims first to compare different S2S calibration methods of the lower limbs in terms of accuracy, repeatability, and usability, and second to specify the impact of the S2S calibration method on lower body kinematics. In Experiment 1, two postures, eight active, and five passive movements, and the use of a device were tested to define lower body segment axes. To isolate the effect of the S2S calibration methods from IMU measures, the IMU measure was mimicked using marker trajectories tracked by an optoelectronic system. The angles between the segment axes obtained with the S2S calibration methods and axes of reference based on an optoelectronic methodology were then compared during a study involving 15 subjects. The results do not reveal unique methods but enable some to be discarded in terms of angle error and repeatability. Following these results, scoring of the methods was proposed in order to help select the most suitable S2S calibration methods given experimental context or subject ability. In Experiment 2, three kinematics obtained with IMUs after S2S calibrations combining the best previous methods, and kinematics obtained with the reference, were compared through root-mean-square error (RMSE), correlation coefficients (CCs), and difference in range of motion (
$\Delta $
ROM) during gait, running, and cycling on an ergocycle. The combination, which included only standing and sitting postures showed significantly lower RMSE (
${p} < 0.01$
) and range of motion differences (
${p} < 0.01$
) in ankle abduction/adduction and plantar/dorsiflexion, respectively, making this combination a good candidate for S2S calibration of the lower body for nonpathological subjects.