Monitoring walking gait parameters with Earables

Introduction
Gait analysis, the study of human locomotion, is
crucial for clinical evaluations and wellness monitor-
ing. Gait analysis has wide-ranging applications in
rehabilitation, sports science, fitness monitoring,
and clinical settings, aiding in injury detection, gait
disorder identification, and early diagnosis of condi-
tions like Parkinson’s disease. Walking gait parame-
ters can be categorized into temporal (e.g., cadence,
stride, stance, and swing time), spatial (e.g., step
length and vertical displacement), and kinetic (e.g.,
ground reaction forces, joint torques). While tem-
poral and spatial parameters are commonly assessed
and associated with overall fitness levels and neuro-
logical conditions, kinetic parameters offer additional
insight into biomechanical determinants of movement
and the loading of anatomical structures.
Outside of the lab, wearables are the main method
for gait assessment. Current methods for wearable
gait analysis involves attaching IMU sensors to the
shoes, ankles and lower back. However, these de-
vices are purpose made with only a singular use case
and can be socially unacceptable or unconformable
to wear. To address these issues, some researchers
have explored using commercial wearables like smart-
watches, however, these have been shown to have lim-
itations in assessing gait metrics.Additionally, wear-
able studies typically focus on specific subsets of gait
parameters for particular applications, lacking a ver-
satile monitoring tool that benefits a broad popula-
tion where a pathology may also be unknown.
Methods
To solve these problems, we propose an ear-worn
wearable, or earable, based walking gait monitoring
system. Compared to other wearable form factors,
earables are desirable as they are widely available
and have dual elements that are located centrally and
symmetrically in the body, allowing symmetric mon-
itoring of both feet. Additionally, the human head
exhibits high stability during walking, which incurs
limited interference and thereby enables accurate gait
monitoring. Additionally, our system can simultane-
ously measure temporal, spatial, and kinetic param-
eters. We also show, for the first time, that earables
are able to reconstruct the vertical ground reaction
force (vGRF) curve and scalar parameters as shown
Figure 1: Example walking vGRF curve and scalars
in fig. 1 Our system leverages IMUs in earphones
to detect gait events and estimate parameters on a
step-by-step basis. It employs lightweight signal pro-
cessing techniques for efficient operation on mobile
devices, compensates for different earphone orienta-
tions, and mitigates the impact of head movements.
Results
We prototyped our system using custom earphones
equipped with 6-axis IMUs. Data was collected from
13 participants who walked in various experimental
settings, including lab-controlled conditions on an in-
strumented treadmill under different walking speeds,
in-the-wild evaluation including free-walking on force
plates, and a stop-and-go scenario, resulting in over
18,000 step samples. Our system achieves an overall
Mean Absolute Percentage Error (MAPE) of 5.1%
for spatio-temporal parameters and 2.0% for kinetic
parameters, and 5.3% on vGRF curve reconstruction.
Compared to existing earable-based gait tracking sys-
tems, our system not only expands the coverage of
gait parameters but also delivers more accurate esti-
mation with better generalisation ability. A full set of
results, benchmarks and on-device performance can
be seen in the published version.