Real-Life Measurement of Tri-Axial Walking Ground Reaction Forces Using Optimal Network of Wearable Inertial Measurement Units

• Published in: IEEE transactions on neural systems and rehabilitation engineering Vol. 26; no. 6; pp. 1243 - 1253
• Main Authors: Shahabpoor, Erfan, Pavic, Aleksandar, Brownjohn, James M. W., Billings, Stephen A., Guo, Ling-Zhong, Bocian, Mateusz
• Format: Journal Article
• Language: English
• Published: United States IEEE 01.06.2018; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Acceleration; Ambulation; Analytical models; Autoregressive moving-average models; Biomarkers; biomechanics; black-box approach; Correlation; Environmental monitoring; Gait; gait monitoring; Legged locomotion; Medical treatment; Motion segmentation; Nonlinear systems; outdoor measurement; Sensors; Sensory systems; System identification; Training; Walking
• ISSN: 1534-4320; 1558-0210; 1558-0210
• DOI: 10.1109/TNSRE.2018.2830976

Monitoring natural human gait in real-life environment is essential in many applications including the quantification of disease progression, and monitoring the effects of treatment and alteration of performance biomarkers in professional sports. Nevertheless, reliable and practical techniques and technologies necessary for continuous real-life monitoring of gait is still not available. This paper explores in detail the correlations between the acceleration of different body segments and walking ground reaction forces GRF(t) in three dimensions and proposes three sensory systems, with one, two, and three inertial measurement units (IMUs), to estimate GRF(t) in the vertical (V), medial-lateral (ML), and anterior-posterior (AP) directions. The nonlinear autoregressive moving average model with exogenous inputs (NARMAX) non-linear system identification method was utilized to identify the optimal location for IMUs on the body for each system. A simple linear model was then proposed to estimate GRF(t) based on the correlation of segmental accelerations with each other. It was found that, for the three-IMU system, the proposed model estimated GRF(t) with average peak-to-peak normalized root mean square error (NRMSE) of 7%, 16%, and 18% in V, AP, and ML directions, respectively. With a simple subject-specific training at the beginning, these errors were reduced to 7%, 13%, and 13% in V, AP, and ML directions, respectively. These results were found favorably comparable with the results of the benchmark NARMAX model, with subject-specific training, with 0% (V), 4% (AP), and 1% (ML) NRMSE difference.