Gait event detection in laboratory and real life settings: Accuracy of ankle and waist sensor based methods

TLDR

Wearable IMU sensors are shifting gait analysis from laboratory to everyday life, yet their event‑detection accuracy has only been validated in controlled lab settings that may not capture real‑world walking patterns. This study assessed the accuracy of two gait‑event detection algorithms—one using shank‑mounted IMUs and the other a waist‑mounted IMU—during free‑living walking. Ten healthy participants performed indoor and outdoor walking tasks, with the algorithms’ outputs compared against pressure‑insole ground truth. The shank algorithm achieved <14 ms error for initial contact, stride, and step times, while final contact and stance times had 28–51 ms errors and variability estimates were 0.09–0.89 %; the waist algorithm missed ~1 % of steps and was less accurate but still acceptable, and both methods performed similarly across conditions, supporting their use in real‑world gait analysis.

Abstract

Wearable sensors technology based on inertial measurement units (IMUs) is leading the transition from laboratory-based gait analysis, to daily life gait monitoring. However, the validity of IMU-based methods for the detection of gait events has only been tested in laboratory settings, which may not reproduce real life walking patterns. The aim of this study was to evaluate the accuracy of two algorithms for the detection of gait events and temporal parameters during free-living walking, one based on two shank-worn inertial sensors, and the other based on one waist-worn sensor. The algorithms were applied to gait data of ten healthy subjects walking both indoor and outdoor, and completing protocols that entailed both straight supervised and free walking in an urban environment. The values obtained from the inertial sensors were compared to pressure insoles data. The shank-based method showed very accurate initial contact, stride time and step time estimation (<14 ms error). Accuracy of final contact timings and stance time was lower (28–51 ms error range). The error of temporal parameter variability estimates was in the range 0.09–0.89%. The waist method failed to detect about 1% of the total steps and performed worse than the shank method, but the temporal parameter estimation was still satisfactory. Both methods showed negligible differences in their accuracy when the different experimental conditions were compared, which suggests their applicability in the analysis of free-living gait.