=Paper=
{{Paper
|id=Vol-2815/CERC2020_paper21
|storemode=property
|title=Exploratory Gait Analysis Using Wearable Technology
|pdfUrl=https://ceur-ws.org/Vol-2815/CERC2020_paper21.pdf
|volume=Vol-2815
|authors=Richard Davies,Paul McCullagh,Huiru Zheng,Joan Condell
|dblpUrl=https://dblp.org/rec/conf/cerc/DaviesMZC20
}}
==Exploratory Gait Analysis Using Wearable Technology==
https://ceur-ws.org/Vol-2815/CERC2020_paper21.pdf
COVID-19 Research and Smart Healthcare
Exploratory Gait Analysis using Wearable Technology
Richard Davies1[0000-0002-2663-3979], Paul McCullagh1[0000-0002-9060-6262], Huiru Zheng1[0000-
0001-7648-8709]
and Joan Condell1[0000-0002-3517-2182]
1
School of Computing, Ulster University, Jordanstown BT37 0QB, Northern Ireland
Abstract. Gait analysis is a rapidly expanding and evolving research area with
application to biomechanics and rehabilitation. Current wearable technology that
can be used to collect gait information is becoming more accessible in terms of
cost and usability compared to lab counterparts which can be expensive and re-
quire training to setup and use. This paper provides an exploratory analysis of
knee angle versus angular velocity of the lower leg for six healthy participants
for at four walking speeds (1.8 km/h, 2.7 km/h, 4.5 km/h, 5.4 km/h) on a treadmill
and walking over the ground at 3.9 km, using the Xsens Inertial Measurement
Unit. These phase portraits provide a rich data source for qualitative comparison.
In future objective gait analysis based on suggested gait parameters can aid with
rehabilitation of dysfunctional gait.
Keywords: gait analysis, wearable technology, inertial motion capture, accel-
erometer, Xsens.
1 Introduction
Gait Analysis (GA) is an area of research that is continually expanding and evolving
across a wide range of domains such as, healthcare, sport science and surveillance.
There are a plethora of medical gait applications such as the evaluation of prosthetics,
assessment of surgical procedures [1] [2], treatments plans, fall risk in the elderly [3]
and assessment of neuropathies [4]. In addition GA has achieved further significance
in the monitoring of elite athletes [5] and identification of individuals for forensic bio-
metric purposes [6] [7].
During the past four decades the measurement and assessment of gait has evolved
rapidly; tools and technology now provide an objective, quantitative evidence-based
approach. Current clinical practice for motor assessment of the lower limb in stroke
survivors is based upon a battery of tests, such as the two-minute walking test, timed-
up and go, berg balance scale, fugl-meyer assessment, motor assessment scale, river-
mead motor assessment of movement, motricity index and stroke rehabilitation assess-
ment of movement. All of the aforementioned motor assessment scales predate the year
1997 and have an average age of 31 years. Although they provide a quantitative score
they are based upon human clinical observation and are subject to inter- and intra-rater
variability. Additionally, the majority of these assessment approaches are not capable
of detecting subtle changes in motor function particularly at the top end of assessment
scales as a ceiling effect often occurs [8].
Copyright © 2020 for this paper by its authors.
Use permitted under Creative Commons License 335 CERC 2020
Attribution 4.0 International (CC BY 4.0).
COVID-19 Research and Smart Healthcare
2
Advances in technology used to measure gait have been instrumental in the evolution
of GA. Biomechanical movement of the human body is complex and therefore effective
GA requires information such as kinematics, ground reaction forces and influence of
muscle activity. Motion Capture (MC) strives to measure kinematic data in an accurate,
valid and unobtrusive manner. There are two competing MC technologies: optical cap-
ture and the use of force plates to measure plantar pressure. Each offers advantages and
disadvantages depending on the context of the application being considered [9] [10].
1.1 Optical motion capture and force plates
Optical Motion Capture (OMC) systems use cameras based upon active or passive
markers to accurately detect the position of body worn markers within a 3-dimensional
space. There are a number of commercial OMC systems available such as Vicon, Qual-
ysis and Codamotion. These systems tend to be very accurate at sensing marker position
to within the sub-millimeter range, however this accuracy heavily relies on the ability
of a researcher to place markers accurately and repeatably. Various protocols exist to
help locate joint centers but these differing conventions can produce a varied set of
results [11]. Due to the complexity involved in camera setup and the configuration of
software, OMC systems require a considerable amount of setup time and a need for
specialised training. Hence OMC systems are more suited to static deployment in a
dedicated gait laboratory, thus impacting upon the information derived as it may not be
representative of gait in a real-world context [12]. OMC systems are expensive and
occupy a static laboratory space which can restrictive for particular applications. How-
ever, they do offer unparalleled accuracy when configured by a trained Biomechanist
and serve as a gold standard or reference point for other less accurate systems.
Traditionally force plates were designed to record single steps with high accuracy
and resolution. Pressure sensing technology has evolved through the incorporation of
this technology into instrumented walkways such as the GAITRite mat facilitating GA
of a sequence of steps [9] [10].
1.2 Wearable technologies and inertial motion capture
Over the last 5 years, there have been significant advances in technologies for inertial
motion capture (IMC) systems; in particular insole pressure sensor recording and the
measurement and wireless transmission of the electromyogram (EMG). Insole pressure
sensing technology has benefited from [13]advancements in microelectronics, wireless
charging, energy harvesting, smaller batteries and low power wireless communication.
These advancements have made insole technology more pervasive, embedding all of
the technology within the insole e.g. Moticon [14]. These developments have paved the
way for wearable technologies to replace gait laboratory equipment in the measurement
of human kinematics, ground reaction forces and muscle activity. These wearable tech-
nologies offer a lower cost, portable, versatile, real-time and highly usable system to
provide rich gait information in free-living environments for clinical GA [15], [16].
Inertial Motion Capture (IMC) systems offer benefits over OMC systems due to
their portability, wearability and decreasing costs. IMC sensor units provide the
CERC 2020 336
COVID-19 Research and Smart Healthcare
3
opportunity for more practical, untethered data capture free from the constraints of an
indoor observation area. Deploying outside a gait laboratory environment can facilitate
diverse spatial settings such as stairs, open space, more natural terrain or other indoor
areas. The setup time is shorter repeatability for measurement of joint angles during
walking [17] is better than OMC systems for both in-day and between-day recording
sessions.
The fundamental assumption of an underlying rigid body such as the human skele-
ton can be violated by the movement of overlying soft tissue known as skin artefact.
This is particularly evident when there is an excess of soft tissue or during highly dy-
namic movements [18]. Skin artefact is a common issue with IMC and OMC, however,
it may be more pronounced with IMC systems as their sensor units are of greater phys-
ical size and mass leading to greater displacement. Skin artefact is an important factor
in the context of clinical research as a higher than average body-fat index is more com-
mon in people suffering from stroke [17]. Ground reaction forces as measured by force
plates provide an alternative or complementary means to perform GA.
2 Methodology
In this study we conducted exploratory GA on healthy participants (n=6) using data
gathered from a wearable IMC system. The GA focuses on a subset of parameters to
assess feasibility, validity and diversity of gait amongst a healthy cohort. Baseline data
were collected to establish a normal set of gait parameters. Factors that may contribute
to variations in the data relate to but are not limited to gender, age, height and weight.
Walking therapies are part of the rehabilitation pathway as defined by the NICE guide-
lines [19]; therefore the focus of this study is on walking on a treadmill and walking
over ground while incorporating a turn.
In a data driven approach the quality and volume of the data being collected plays a
vital role in the capability of a computational model to provide accurate, objective and
repeatable assessment of gait. Therefore, it is import to control the recording of activi-
ties by applying a consistent clinical protocol. Although the number of participants is
low (n=6) the size of the dataset can still be adequate to provide sufficient quality of
data as the number of steps can reach 2,400. In addition each step can provide further
information angle, velocity for 7 locations on the lower body.
Participants wore an IMC system and a pair of smart insoles to collect kinematic and
ground reaction forces during walking activities. Participants were required to complete
a short anonymous questionnaire that provided information on their age, gender and
weight.
Inertial Motion Capture Configuration. The IMC system was configured to investi-
gate lower limb which involved donning 7 Inertial Measurement Units (IMUs) using a
velcro based strapping system as shown in Fig. 1. The IMUs were attached to the pelvis
(sacrum), left/right upper leg (thigh), left/right lower leg (shank) and left/right foot. A
number of anatomical measurements (body height, shoulder width, arm span, hip
height, hip width, knee height, ankle height, foot size, shoe sole height) were taken from
337 CERC 2020
COVID-19 Research and Smart Healthcare
4
the participant to help complete the calibration. The IMC was re-calibrated before each
recording session.
Fig. 1. Xsens inertial capture system showing sensor positions for foot, lower and upper limb.
Participants. A group of 6 healthy adults participated in the study. A summary of de-
mographic information is provided in Table 1.
Table 1. Demographic information of participants
Participant ID Gender Age Height (cm) Weight (Kg)
1 Male 61 179 68.7
2 Female 52 158 N/A
3 Male 26 186 N/A
4 Male 71 172 73.0
5 Female 71 164 64.0
6 Male N/A 177 N/A
Initial feasibility testing was conducted to evaluate at an observational level that the
wearable systems were fit for purpose in terms of robustness, reliability, usability, com-
fort, repeatability and set-up time. Walking activities have been designed to include
walking at differing speeds, walking on a treadmill versus over the ground. The study
also included turns as this is an important constituent of GA.
Given that the potential use case scenario for this research will be typically elderly
post-stroke survivors some slower walking speeds were included. Participants per-
formed two walking activities. The first involved walking indoors over the ground on
a flat smooth surface within a gait laboratory environment for a distance of 10m and
turning. This activity was repeated for 2.5 mins at a comfortable walking speed, self-
selected by the participant. The second activity required participants to walk for 2.5
minutes at 4 different speeds on a treadmill with zero incline for a total walking time
of 10 minutes. The four speeds using the treadmill were: very slow (0.5m/s), slow
(0.75m/s), medium (a comfortable speed self-selected by the participant, either 1m/s or
CERC 2020 338
COVID-19 Research and Smart Healthcare
5
1.25m/s) and fast (1.5m/s). The fastest walking speed test is similar to the 2-minute
walking test which is a clinical assessment.
3 Results
The Xsens system has already been well validated against OMC systems and has been
reported to have a coefficient of multiple correlation > 0.96 for all joints during flex-
ion/extension for level walking activities [20]. However, configuration, calibration and
positioning sensors can have varying effects on the quality of the data collected. Re-
peatability was explored for in-day testing of the Xsens without doffing/donning the
sensors.
The two phase portraits shown in Fig.2 are highly correlated and present knee angle
against angular velocity of the lower limb for participant 1. These phase portraits show
the dynamic nature of the knee during walking on a treadmill at a 5.4 m/s for 2 minutes.
A single gait cycle is represented by one phase which can be seen as a closed loop. The
phases are plotted on top of each other as each gait cycle is repeated, it shows high
levels of correlation but with some dynamic and chaotic variations. The variation be-
tween gait cycles in the first test can be seen in red while the blue lines show the vari-
ation of gait cycles in the second test. Since both tests were recorded within 30 minutes
and under the same conditions the variation which is expected to be minimal between
walks can be observed by comparing red and blue lines.
Fig. 2. A highly correlated phase portrait of knee angle versus angular velocity of the lower
right leg for Participant 1 during two separate tests while walking on treadmill at 5.4 m/s for 2
minutes.
To provide a statistical measure of correlation an average gait cycle was computed for
both walks and a correlation coefficient calculated by comparing both average gait
339 CERC 2020
COVID-19 Research and Smart Healthcare
6
cycles. The average phase portraits can be seen in Fig. 3, these are highly correlated as
expected (r=0.9993), this is the Pearson correlation coefficient as reported by
MATLABs 2-dimensional correlation function.
Fig. 3. A highly correlated (r=0.9993) phase portrait, it represents an average gait cycle from
two walks by participant 1.
The next stage of GA was to compare all of the walking activities for participants across
a number of different walking speeds; this involved 2 minute treadmill walks at 1.8
km/h, 2.7 km/h, 3.6 km/h or 4.5 km/h and 5.4 km/h and a final 2 minute walk over the
ground for 10 metres with a turn. The phase portraits of these walking activities can be
seen in Fig. 4 for participant 1. Initial observations show correlation between walking
speed and the area enclosed within the phase portraits. A greater range of motion and
increased angular velocity should result in an increased area within the curves. Addi-
tionally, as the speed increases the variability in the phase portraits reduces to produce
a more rhythmic and stable gait cycle, this is particularly for true for walking activities
on the treadmill. It seems that this effect is caused by a combination of the treadmill
and higher walking speeds.
Fig. 4. Phase portraits of knee angle versus angular velocity of the lower right leg for Partici-
pant 1 at four walking speeds (1.8 km/h, 2.7 km/h, 4.5 km/h, 5.4 km/h) on a treadmill walking
over the ground at 3.9 km/h.
CERC 2020 340
COVID-19 Research and Smart Healthcare
7
To quantify the relationship between the area of the phase portraits and speed an aver-
age phase portrait was computed for each walk. These were then used as a basis to
calculate an average area of each phase portrait and to plot these against speed to quan-
tify any relationship and how it may change across the cohort. Observation of Fig. 5
shows a common pattern of increased area equating to increased speed. Participants 4
and 5 were both in their early seventies and yet it is interesting that they were able to
maintain walking speeds of 1.8 km/h, 2.7km/h and 3.3km/h with a reduced area, this
may imply a greater sense of control by reducing the stride length.
Fig. 5. Walking activity for six participants showing average phase plot area against speed.
Comparing treadmill walking activities against over the ground walking in Fig. 4 shows
that the variation in knee angle is more apparent while walking over ground. As the
walking over the ground activity included turning 180 degrees every 10 metres there
are a significant amount of turns (n~=16) within a 2 minute period. Therefore it makes
it more difficult to attribute the increased knee angle variation as a direct result of walk-
ing over the ground. The second noticeable feature of the phase portraits for walking
over the ground is that there is a mirroring effect which results in two prominent distinct
phases. These are a direct result of walking in two opposite directions and may be com-
bined into a single phase if it is possible to adjust the data to accommodate walking
direction. This would be a useful analysis feature as it would allow all walking activities
to be analyzed and compared irrespective of walking direction.
4 Future Work
This paper provided a demonstration that repeatable and interpretable gait analysis is
possible using wearable IMC technologies. Phase cycles and repeatability were as-
sessed by observation and quantitative measurement. Further work will be conducted
341 CERC 2020
COVID-19 Research and Smart Healthcare
8
with more healthy participants and incorporate between-day and inter-rater reliability.
There is a significant opportunity to quantify GA through the measurement of spatial,
temporal, spatiotemporal and other phasic parameters, as listed in Table 2. Additional
features that can be used in assessing the repeatability of normal gait can be derived
from the normative dataset. Using these measurements and features will facilitate gait
modelling for a healthy population. A gait model for a healthy population can be used
to provide a reference to any new gait information that should be compared from a
disease specific cohort such as stroke survivors.
There are a number of use cases where gait analysis can be beneficial for both the cli-
nician and patient such as Parkinson’s disease, cerebral palsy, lower-limb osteoarthritis,
post-stroke and diabetic neuropathy. The authors propose to develop a computational
gait model based on intelligent data analysis using non-linear techniques to provide an
objective and quantifiable assessment of pathological gait for post-stroke that is accu-
rate, robust and repeatable.
Table 2. List of gait features for future work.
Spatial Temporal Spatiotemporal Phasic
Step Length (cm) Cadence Gait Speed (m/s) Stance Time
(steps/min) (%GC)
Stride Length Step Time (s) Stride Speed (m/s) Swing Time
(cm) (%GC)
Step Width (cm) Stride Time (s) Stride variability SST (%GC)
Step Height (cm) Stance Time (s) Smoothness DST (%GC)
Knee Angle (°) Swing Time (s) Centre of Pressure
Hip Angle (°) Single Support
Time (s)
Ankle Angle (°) Double Support
Time (s)
5 Conclusion
Gait analysis of the normal population and of different pathologies is an area of research
that is expanding rapidly. There are a number of competing technologies that can pro-
vide gait information, two such competing sets are research gait lab technology and
wearable technology. The former tends to be more expensive, less flexible and with
longer setup times often requiring specialised training. With recent advances, wearable
technology can offer a cheaper, more accessible, less restrictive and easier to use option
without comprising on the accuracy or quality of the information. This is particularly
true of recent advances of IMC systems.
The Xsens IMC system captured kinematic data from walking. Due to the explora-
tory nature of this study only the dynamic nature of knee angle during walking was
considered; the gait variation across a number of walking speeds on a treadmill and
walking over the ground and gait variation across the population were assessed. Future
CERC 2020 342
COVID-19 Research and Smart Healthcare
9
research aims to build a computational model that can be used to assess a user’s gait
during ambulation. A large set of features will be generated to serve as input to the
model and as such can be configured in multiple ways via feature selection to ascertain
the optimal model and as a result what are the optimal features, technologies and sen-
sors. There is a significant body of research to suggest that spatial temporal gait param-
eters provide such a feature set [21], [22], [23].
6 Acknowledgements
The authors would like to acknowledge the School of Computing at Ulster University
and SENDoc (www. Sendocnpa.com) project in the ongoing support of this doctoral
research.
References
1. A. Heikkilä, N. Sevander-Kreus, A. Häkkinen, M. Vuorenmaa, P. Salo, P. Konsta and
J. Ylinen, "Effect of total knee replacement surgery and postoperative 12 month home
exercise program on gait parameters," Gait & Poature, vol. 53, pp. 92-97, 2017.
2. P. Ornettia, J.-F. Maillefert, D. Laroche, C. Morisset, M. Dougados and L. Gossec,
"Gait analysis as a quantifiable outcome measure in hip or knee osteoarthritis: A
systematic review," Joint Bone Spine, vol. 77, pp. 421-425, 2010.
3. M. Zagoa, C. Sforza, D. R. Bonardi, E. E. Guffanti and M. Galli, "Gait analysis in
patients with chronic obstructive pulmonary disease: a systematic review," Gait &
Posture, vol. 61, pp. 408-415, 2018.
4. M. E. Fernando, R. G. Crowther, P. A. Lazzarini, K. S. Sangla, P. Buttner and J.
Golledge, "Gait parameters of people with diabetes-related neuropathic plantar foot
ulcers," Clinical Biomechanics, vol. 37, pp. 98-107, 2016.
5. B. Hanley, C. B. Tucker and A. Bissas, "Differences between motion capture and video
analysis systems in calculating knee angles in elite-standard race walking," Journal of
Sport Science, vol. 36, no. 11, pp. 1250-1255, 2018.
6. I. Bouchrika, M. Goffredo, J. Carter and N. Mark, "On Using Gait in Forensic
Biometrics," Journal of Forensic Sciences, vol. 56, no. 4, pp. 882-889, 2011.
7. Y. He, J. Zhang, H. Shan and L. Wang, "Multi-Task GANs for View-Specific Feature
Learning in Gait Recognition," IEEE Transactions on Information Forensics and
Security, vol. 14, no. 1, pp. 102-113, 2019.
8. D. J. Gladstone, C. J. Danells and S. E. Black, "The Fugl-Meyer Assessment of Motor
Recovery after Stroke: A Critical Review of Its Measurement Properties,"
Neurorehabilitation and Neural Repair, vol. 16, no. 3, pp. 232-240, 2002.
9. S. S. Kuys, S. G. Brauer and L. Ada, "Test-retest reliability of the GAITRite system in
people with stroke undergoing rehabilitation," Disability and Rehabilitation, vol. 33,
no. 19-20, pp. 1848-1853, 2020.
343 CERC 2020
COVID-19 Research and Smart Healthcare
10
10. A. Scully and D. Tan, "Within-Assessment Variability of the GAITRite System in
People With Stroke," Physical Medicine & Rehabilititation, vol. 98, no. 6, pp. 446-
449, 2019.
11. A. Ferrari, M. G. Benedetti, E. Pavan, C. Frigo, D. Bettinelli, M. Rabuffetti, P. Crenna
and A. Leardini, "Quantitative comparision of five current protocols in gait analysis,"
Gait Posture, vol. 28, no. 2, pp. 207-216, 2008.
12. A. Ali, K. Sundaraj, B. Ahmad, N. Ahamed and I. Anamul, "Gait disorder rehabilitation
using vision and non-vision based sensors: A systematic review," Bosnian journal of
basic medical sciences, vol. 12, no. 3, pp. 193-202, 2012.
13. T. P. Kakarla, K. A. Varma, P. S. P. Jayaraj Joseph and M. Sivaprakasam, "Accuracy
Enhancement of Total Force by Capacitive Insoles," in IEEE International Symposium
on Medical Measurements and Applications, Istanbul, 2019.
14. M. S. Oerbekke, M. J. Stukstette, K. Schütte, R. A. de Bie, M. F. Pisters and B.
Vanwanseele, "Concurrent validity and reliability of wireless instrumented insoles
measuring postural balance and temporal gait parameters," Gait & Posture, vol. 51, pp.
116-124, 2017.
15. J. Lee, S. Y. Shin, G. Ghorpade, T. Akbas and J. Sulzer, "Sensitivity comparison of
inertial to optical motion capture during gait: implications for tracking recovery," in
IEEE 16th Internal Conference on Rehabilitation Robotics, Toronto, 2019.
16. C. Cao, M. I. Rodríguez-Ferradas, A. Cazón-Martin and P. Morer, "Development of a
Low-Cost Wearable Prevention System for Musculoskeletal Disorders Using Inertial
Measurement Unit Systems," in Advances on Mechanics, Design Engineering and
Manufacturing II. Lecture Notes in Mechanical Engineering, Cham, Springer, 2019.
17. T. Cloete and C. Scheffer, "Quantification of human motion: gait analysis—benefits
and limitations to its application to clinical problems," in Quantification of human
motion: gait analysis—benefits and limitations to its application to clinical problems,
Quantification of human motion: gait analysis—benefits and limitations to its
application to clinical problems, 2010.
18. A. Cappozzo, F. Catani, A. Leardini, M. G. Benedetti and D. U. Croce, "Position and
orientation in space of bones during movement: experimental artefacts," Clinical
Biomechanics, vol. 11, no. 2, pp. 90-100, 1996.
19. "National Institute for Health and Care Excellence Stroke rehabilitation: long-term
rehabilitation after stroke. Clinical Guideline 162," NICE, 2013. [Online]. Available:
www.nice.org.uk/guidance/cg162. [Accessed 31 08 2020].
20. J.-T. Zhang, A. C. Novak, B. Brouwer and Q. Li, "Concurrent validation of Xsens
{MVN} measurement of lower limb joint angular kinematics," Physiological
Measurement, vol. 34, no. 8, pp. 63-69, 2013.
21. J. H. Hollman, M. K. Watkins, A. C. Imhoff, C. E. Braun, K. A. Akervik and D. K.
Ness, "A comparison of variability in spatiotemporal gait parameters between treadmill
and overground walking conditions," Gait & Posture, vol. 43, pp. 204-209, 206.
22. O. Beauchet, G. Allali, H. Sekhon, J. Verghese, S. Guilain, J.-P. Steinmetz, R. W.
Kressig, J. M. Barden, T. Szturm, C. P. Launay, Grenier, L. Bherer, T. Liu-Ambrose,
V. L. Chester and M. L. Callisaya, "Guidelines for Assessment of Gait and Reference
Values for Spatiotemporal Gait Parameters in Older Adults: The Biomathics and
Canadian Gait Consortiums Initiative," Frontiers in Human Neuroscience, vol. 11, p.
353, 2017.
CERC 2020 344
COVID-19 Research and Smart Healthcare
11
23. R. Briggs, D. Carey, P. Claffey, T. McNicholas, O. Donoghue and S. P. Kennelly, "Do
Differences in Spatiotemporal Gait Parameters Predict the Risk of Developing
Depression in Later Life?," Journal of the American Geriatrics Society, vol. 67, no. 5,
pp. 1050-1056, 2019.
345 CERC 2020