Ground reaction forces (GRF) and joint kinematics are critical parameters for gait analysis. The conventional approach to acquiring these data is to use force plates installed under the walkway and a stationary motion capture system. These instruments make real-world analysis difficult and incur high costs. One possible solution is to record kinematics with wearable sensors and to apply machine learning models to predict the GRF. However, a protocol that suits online applications and takes greater advantage of portable measurements has been underinvestigated. This study employed Extreme Gradient Boosting (XGBoost) to estimate within-subject three-dimensional (3D) GRF using five lower limb joint angles: hip flexion-extension, adduction-abduction and internal-external rotation, knee flexion-extension, and ankle plantar-dorsi flexion. These joint angles were computed from publicly available data, captured with a camera system (chosen as a benchmark), from six participants. For each subject, we used four-fold cross-validation to assess the estimator performance. The results showed that the model performed the best for anterioposterior direction, followed by vertical direction, then mediolateral direction, with median average 푹ퟐ values of 0.96, 0.90 and 0.64, respectively. This study demonstrated 3D GRF prediction based on kinematic features that could be obtained in an online manner. This simple protocol may improve the practicality of carrying out online analyses outside the laboratory with a small number of wearable sensors such as inertial measurement units.
|Otsikko||Proceedings of the 5th International Conference on NeuroRehabilitation, ICNR2020|
|Tila||Hyväksytty/In press - 2020|
|OKM-julkaisutyyppi||A4 Artikkeli konferenssijulkaisuussa|
|Tapahtuma||International Conference on NeuroRehabilitation - Virtual, Online|
Kesto: 13 lokakuuta 2020 → 16 lokakuuta 2020
|Conference||International Conference on NeuroRehabilitation|
|Ajanjakso||13/10/2020 → 16/10/2020|