Dynamic Gait Stability Estimated Using One or Two Inertial Measurement Units Worn on the Human Body

TL;DR

This study shows that using two wearable sensors can accurately estimate walking stability, which could help identify people at risk of falling.

Contribution

The study identifies optimal body locations for two IMUs to predict gait stability with accuracy comparable to gold-standard motion capture.

Findings

Two IMUs placed at optimal body locations achieved mean errors of ~30 mm and 11 mm in anterior and mediolateral MoS.

IMU-based MoS estimation is accurate enough for classifying individuals at high fall risk.

Principal motion analysis was effective for predicting MoS from multidimensional IMU data.

Abstract

The margin of stability (MoS) is a metric used to assess dynamic postural stability during walking. Although MoS is typically computed from optical motion capture data, previous studies have shown that MoS can be approximated from six-axis kinematic signals—linear acceleration and angular velocity—measured by inertial measurement units (IMUs). With IMU-equipped devices such as smartphones and smartwatches becoming widespread, it is increasingly common for individuals to carry two or more such devices in daily life. This study aimed to identify combinations of two body locations that most effectively predict MoS. IMU sensors were attached to ten body locations while participants walked on a treadmill. Principal motion analysis, a reductive regression method for multidimensional time-series data, was employed for MoS prediction, and cross-validation was used for reliable model evaluation. Appropriate combinations of two IMU sensors achieved mean errors of approximately 30 mm and 11 mm in anterior and mediolateral MoS, respectively, compared with reference values derived from optical motion capture. These errors were comparable to the intrinsic standard deviations of MoS, suggesting that IMU-based MoS estimation is sufficiently accurate for the classification of individuals at high fall risk.

Figures