Comparative Analysis of Inertial Sensor to Optical Motion Capture System Performance in Push-Pull Exertion Postures

Abstract

This study examined interactions between inertial sensor (IS) performance and physical task demand on posture kinematics in a two-handed force exertion task. Fifteen male individuals participated in a laboratory experiment that involved exerting a two-handed isometric horizontal force on an instrumented height-adjustable handle. Physical task demand was operationalized by manipulating vertical handle height, target force magnitude, and force direction. These factors were hypothesized to influence average estimates of torso flexion angle measured using inertial sensors and an optical motion capture (MC) system, as well as the root mean squared errors (RMSE) between instrumentation computed over a 3s interval of the force exertion task. Results indicate that lower handle heights and higher target force levels were associated with increased torso and pelvic flexion in both, push and pull exertions. Torso flexion angle estimates obtained from IS and MC did not differ significantly. However, RMSE increased with target force intensity suggesting potential interactive effects between measurement error and physical task demand.

Full text

Comparative Analysis of Inertial Sensor to Optical Motion Capture
System Performance in Push-Pull Exertion Postures
Sol Lim, Andrea Case, Clive D’Souza
Center for Ergonomics, Department of Industrial and Operations Engineering,
University of Michigan, Ann Arbor
This study examined interactions between inertial sensor (IS) performance and physical task de-
mand on posture kinematics in a two-handed force exertion task. Fifteen male individuals partici-
pated in a laboratory experiment that involved exerting a two-handed isometric horizontal force
on an instrumented height-adjustable handle. Physical task demand was operationalized by ma-
nipulating vertical handle height, target force magnitude, and force direction. These factors were
hypothesized to influence average estimates of torso flexion angle measured using inertial sensors
and an optical motion capture (MC) system, as well as the root mean squared errors (RMSE) be-
tween instrumentation computed over a 3s interval of the force exertion task. Results indicate that
lower handle heights and higher target force levels were associated with increased torso and pel-
vic flexion in both, push and pull exertions. Torso flexion angle estimates obtained from IS and
MC did not differ significantly. However, RMSE increased with target force intensity suggesting
potential interactive effects between measurement error and physical task demand.
INTRODUCTION
Prolonged work in awkward constrained postures
and high force exertions are known risk factors for mus-
culoskeletal disorders in manual work settings (da Costa
& Vieira, 2010). Overexertion injuries have been associ-
ated with repeated and long term bending and twisting of
trunk at the waist (NIOSH, 2009). Estimating the physi-
cal workload experienced by workers in the workplace
necessitates reliable and accurate methods to measure
worker postures and work intensities in situ.
Low-cost, wearable inertial sensors (IS) provide a
useful tool for quantifying postures and physical expo-
sures in field settings. These small and inexpensive sen-
sors comprising of integrated accelerometers, gyro-
scopes and magnetometers have been used in various
domains such as gait analysis (Charry et al., 2009) and
activity monitoring (Oshima et al., 2011; Ravi et al.,
2005).
Prior studies have investigated the accuracy of IS
compared to optoelectronic motion capture (MC) sys-
tems (e.g., Faber et al., 2013; Godwin et al., 2009; Cutti
et al., 2008). These studies have largely considered
measurement differences to be independent of body pos-
ture, which is a function of worker anthropometry and
physical task demand. A systematic investigation of IS
performance as a function of occupationally-relevant
task variables and task demands is a necessary step to-
wards determining task conditions in which IS might
outperform or underperform relative to conventional lab-
based instrumentation methods.
We conducted a laboratory experiment to compare
the performance of a low-cost IS to a conventional MC
system in simulated static and dynamic work tasks under
levels of physical task demand. Preliminary analysis and
findings of the two-handed isometric pushing and pull-
ing tasks are presented here.
The specific study aim was to quantify differences
in IS and MC estimates of torso and pelvis kinematics
relative to levels of physical task demand in a two-
handed isometric horizontal hand force exertion task.
Task demand was operationalized by the magnitude (in-
tensity), direction and height of force application. We
hypothesized torso and pelvic flexion to change system-
atically between task demands but not between instru-
mentation methods (IS vs. MC as reference).
METHODS
Study Participants
The study recruited fifteen healthy right-handed
males (18-35 years old) from the university population.
Age and gender restriction were applied to minimize
variability in self-selected task postures. Data from three
participants were excluded from the analysis due to in-
strument error. The resultant sample (n = 12) had a mean
(SD) age of 24.21 (3.98) years, height of 176.52 (4.57)
cm, and weight of 69.79 (9.0) kg. Participants provided
written informed consent prior to study participation and
were screened for pre-existing back injuries or chronic
pain using a questionnaire. The study was approved by
the University’s Health Sciences and Behavioral Scienc-
es Institutional Review Board.
Not subject to U.S. copyright restrictions. DOI 10.1177/1541931213601224
Proceedings of the Human Factors and Ergonomics Society 2016 Annual Meeting 970

Experiment Procedure
A laboratory experiment was conducted that in-
volved participants exerting a two-handed isometric hor-
izontal force on a height-adjustable handle instrumented
with a 6-axis load cell. Participants exerted a hand force
to match and maintain a required target force level
(±5%) for a continuous 3s interval in 18 counterbalanced
tasks conditions (Table 1) characterized by vertical han-
dle height (at the hip, mid-point, and shoulder height),
force intensity (low, medium, and high), and force direc-
tion (push vs. pull).
Table 1. Target force intensity levels for the two-handed push
and pull task presented as a percentage of two-handed maxi-
mum push exertion force measured at hip height. Pull vs. push
ratio was determined using reference MVE values provided by
Chaffin & Andres (1983).
Target force levels were normalized to each partici-
pant and set proportional to their maximum voluntary
exertion (MVE) force in a two-handed push with the
handle at hip height. The MVE was measured at the start
of the experiment. Participants received verbal instruc-
tions to ramp up their force exertion to a maximum over
2s interval and maintain it for a 3s interval (Stobbe,
1982). The average of two repetitions was recorded as
the MVE. The resultant study sample had a hip height of
100.6 (3.59) cm. and a push MVE of 455.98 (126.05) N.
For the experiment trials, participants stood with
their feet positioned in a split stance with the dominant
foot leading and placed within 50 cm from the handle.
Participants were free to self-select the non-dominant
foot position in each task trial. A digital display monitor
located anteriorly at eye level provided real-time feed-
back of the exerted force and target force range with a 3s
countdown clock, and was reset for each task condition.
Failure to produce a sustained force level within the tar-
get range for a continuous 3s resulted in the clock being
reset and the trial repeated.
Instrumentation
Measurement instrumentation comprised of a com-
mercial data-logging IS devices (YEI Technology, Inc.)
and a passive optical MC system (Qualisys Inc.). Two
sensors were used in this analysis and were attached us-
ing Velcro straps to the participant’s torso at sixth tho-
racic (T6) vertebra for tracking 3D orientations of the
upper torso, and at the low-back (L5/S1) for measuring
pelvic orientation. Figure 1 (left panel) shows the ana-
tomical locations for the IS and MC markers. MC mark-
er triads attached to the IS devices were used to track
participant movements during the experiment trials. Ad-
ditional MC markers attached to the participant’s acro-
mion process, cervicale (C7) and hip (greater trochanter)
were used in a static reference pose (T-pose) measure-
ment to map the marker triads to the upper torso and
pelvis segment orientations.
Figure 1. Experimental setup (left) showing anatomical reference location for the inertial sensors (rectangles, 2 nos.) and optical MC
marker triads (circles) located at the upper torso (T6) and pelvis (L5/S1). MC markers at the acromion process, cervicale (C7) and hip
were used in a static pose measurement to map the marker triads to the upper torso and pelvis segment orientations. During the task
trials, posture variables (right panel) comprising torso flexion ‘θ’ and pelvis flexion ‘ϕ’ angles were calculated relative to the upright
standing posture for each instrumentation system.
Proceedings of the Human Factors and Ergonomics Society 2016 Annual Meeting 971

Data Processing
During the experiment, the IS devices recorded tri-
axial accelerometer, gyroscope, and magnetometer data
at 100 Hz sampling frequency. MC data was collected at
a sampling frequency of 50-Hz. Both IS and MC data
were filtered using a second-order low-pass zero-lag
Butterworth filter with a 6-Hz cut-off frequency. MC
data were then up-sampled to 100-Hz and synchronized
with the IS data. Three dimensional segment orientations
using IS data (roll and pitch were used in this analysis)
were computed using a custom algorithm implemented
in MATLAB.
Dependent Variables
Dependent measures comprised of the posture varia-
bles, torso flexion (TF, ‘θ’) and pelvis flexion (PF, ‘ϕ’)
angles, calculated separately from the IS and MC data
for each trial. In interest of brevity we have limited our
analysis to torso flexion. The following values were
computed:
(i) Mean flexion angles over the 3s task duration
(ii) Standard deviations (SD) over the 3s task duration
(iii) Root mean squared errors (RMSE) between IS and
MC for the 3s task duration.
Data Analysis
Hierarchical mixed effects models (Snijder and Bos-
ker, 2002) implemented using SPSS v. 23 (IBM Inc.)
were used to analyze the effects of within-subject task
variables, viz., handle height (hip, mid, shoulder), force
intensity (low, mid, high) and force direction (push vs.
pull) and between instrumentation (IS vs. MC) on TF.
Two different models were used to evaluate measure-
ment error across instrumentation (IS vs. MC). The first
model investigated difference in mean TF angles by task
variables and instrumentation. The second model inves-
tigated for differences in RMSE by task variables. A
0.05 nominal significance level was used in the analyses.
All main and two-way interaction terms were included in
the analysis. Post hoc paired comparisons of significant
terms (p < 0.05) used the Bonferroni adjustment.
RESULTS
Table 2 summarizes the mean (SD) angle values for
TF by instrumentation and the mean (SD) RMSE values
between instrumentation stratified by task variables.
Overall trends in mean TF values show that lower han-
dle heights and higher target force levels are associated
with an increase in TF. On average, TF was greater in
push compared pull exertions.
Overall, the RMSE between IS and MC for 3s of TF
ranged between 1.80 – 3.69 degrees (Table 2). RMSE
during the 3s task duration tended to increase with lower
handle height, higher force intensity, and push task vari-
ables.
Analysis of Mean TF Angle
Mixed effects analysis of TF angles (Table 3)
showed no significant effect of instrumentation (IS vs.
MC) on the 3s-averaged TF values. This result suggests
that when averaged over 3s, IS and MC provided TF
angles estimates that were consistent and independent of
task variables.
Table 2. Mean (SD) values for torso flexion angle (degrees) over the 3s task duration derived from Motion Capture (MC) and Inertial
sensors (IS) stratified by task variables, along with the root mean squared errors (RMSE mean and SD) between MC and IS.
Handle
Height
Force
Intensity
Force Direction
Push
Pull
Mean (SD)
RMSE (SD)
MC vs. IS
Mean (SD)
RMSE (SD)
MC vs. IS
MC
IS
MC
IS
Shoulder
Low
14.47 (6.13)
14.19 (5.55)
1.94 (2.38)
8.60 (7.84)
8.85 (7.12)
1.85 (2.36)
Medium
18.07 (7.70)
17.37 (7.27)
2.19 (2.41)
7.55 (7.76)
7.94 (7.21)
1.89 (2.45)
High
18.66 (8.43)
18.49 (7.98)
2.13 (2.41)
16.73 (13.55)
16.50 (12.30)
2.26 (2.48)
Mid
Low
22.93 (6.06)
22.17 (6.0)
2.22 (2.41)
13.37 (10.63)
13.39 (10.19)
1.80 (2.33)
Medium
26.19 (6.14)
25.36 (6.0)
2.38 (2.35)
21.80 (12.02)
20.69 (11.17)
2.02 (2.55)
High
26.93 (6.70)
26.37 (6.33)
2.49 (2.20)
25.88 (14.85)
24.66 (14.49)
2.23 (2.25)
Hip
Low
39.90 (10.28)
38.40 (10.15)
2.63 (2.24)
29.95 (28.58)
28.58 (12.44)
2.49 (2.52)
Medium
42.12 (11.30)
40.26 (9.76)
3.36 (2.20)
38.62 (9.17)
36.28 (8.40)
3.09 (2.25)
High
44.98 (14.27)
43.12 (13.12)
3.69 (2.19)
42.41 (12.43)
40.25 (11.61)
3.07 (2.23)
Proceedings of the Human Factors and Ergonomics Society 2016 Annual Meeting 972

Significant influences of handle height, force inten-
sity, force direction, and an interaction between force
intensity and direction were observed (Table 3). Post-
hoc tests showed significantly greater TF angles at lower
handle heights, viz., shoulder to mid handle height: t = -
10.01, p < 0.001, shoulder to hip height: t = -28.83, p <
0.001, and mid to hip height: t = -18.74, p < 0.001.
Force intensity (high-mid: t = 4.05, p < 0.001, high-
low: t = 8.78, p < 0.001, med-low: t = 4.66, p < 0.001)
and force direction (push vs. pull: t = 8.03, p < 0.001)
were the significant factors, implying significantly
greater TF with increasing force intensity also when
pushing.
Table 3. Summary results from the mixed-effects analysis of
mean TF angle as a function of task variables and instrumenta-
tion. (* indicates significant effect at p < 0.05)
Source
Torso Flexion Angle
Intercept
F(1, 24) = 349.05, p < 0.001 *
Handle Height
F(2, 404) = 427.80, p < 0.001 *
Force Intensity
F (2, 404) = 38.57, p < 0.001 *
Force Direction
F (1, 404) = 64.48, p < 0.001 *
Instrumentation (IS vs. MC)
F (1, 24) = 0.12, p = 0.738
Handle Height * Force Intensity
F (4, 404) = 1.42, p = 0.225
Handle Height * Force Direction
F (2, 404) = 0.11, p = 0.90
Force Intensity * Force Direction
F (2, 404) = 7.10, p < 0.001 *
Handle Height * Instrumentation
F (2, 404) = 0.51, p = 0.60
Force Intensity * Instrumentation
F (2, 404) = 0.046, p = 0.955
Force Direction * Instrumentation
F (2, 404) = 0.013, p = 0.909
Analysis of TF Angle RMSE
Mixed-effects analysis of RMSE between IS vs. MC
estimates of TF angles showed significant effect of han-
dle height, force intensity, and force direction (Table 4).
No interactions between factors were observed. Post-hoc
tests showed significant increase in RMSE for handle
heights at the hip vs. shoulder (t = 7.25, p < 0.001), high
vs. mid force intensity levels (t = 3.56, p < 0.001), and in
push vs. pull force exertions (t = 2.18, p = 0.031).
Table 4. Summary results from the mixed-effects analysis of
torso flexion RMSE as a function of task variables. (* indi-
cates significant effect at p < 0.05)
Source
Torso Flexion RMSE
Intercept
F(1, 12) = 16.24, p = 0.002 *
Handle Height
F(2, 201) = 30.63, p < 0.001 *
Force Intensity
F (2, 201) = 6.54, p = 0.002 *
Force Direction
F (1, 201) = 4.71, p = 0.031 *
Handle Height * Force Intensity
F (4, 201) = 0.94, p = 0.441
Handle Height * Force Direction
F (2, 201) = 0.511, p = 0.601
Force Intensity * Force Direction
F (2, 201) = 0.02, p = 0.980
DISCUSSION
This study examined interactions between inertial
sensor performance and physical task demand on posture
kinematics in a hand force exertion task. Findings
demonstrate a systematic influence of physical task de-
mands on torso flexion.
Regarding the performance of IS, no significant dif-
ference between IS and MC angle estimates were ob-
served when posture variables were averaged over the
task duration. However, comparisons of RMSE during
the exertion task by condition indicated greater error
magnitudes in conditions representing higher force in-
tensity, lower handle heights, and in push exertions.
While the output force was relatively isometric (target ±
5%), it is likely that body posture, specifically TF did
vary during the 3s task duration, either volitionally or
due to tremor in conditions of high force intensity (75%
MVE). Differences in the response characteristics of IS
compared to optical motion capture in recording such
movements may have contributed to greater RMSE.
Further, the accuracy of IS when measuring changes
in three dimensional angles has some limitations, name-
ly, fluctuation from accelerometer and drift errors in gy-
roscope–based measurements (Welch & Foxlin, 2002).
Godwin et al. (2009) reported RMSE in IS vs. MC esti-
mates ranging between 3 and 15 degrees when studying
human arm reaching movement, compared to a RMSE
of 3.5 degrees in simple dynamic pendulum movements.
Our preliminary study suggests that inertial sensor
measurement errors differ based on specific posture an-
gles being compared (which is a function of worker an-
thropometry and physical task demands) and the choice
of data aggregation method (i.e., whether the data is be-
ing averaged or not). Similar angle estimates were ob-
tained when averaged over the task duration, though the
variability in measurements across conditions differed.
The present study was limited to isometric two-
handed exertions. Ongoing work to include additional
body segments and dynamic work tasks will help pro-
vide a more comprehensive understanding of IS perfor-
mance for ergonomic analysis.
ACKNOWLEDGEMENT
This study was supported by a pilot grant to the first
author by the U-M Center for Occupational Health and
Safety Engineering (COHSE) through the training grant
T42-OH008455 from the National Institute for Occupa-
tional Safety and Health, Centers for Disease Control
and Prevention.
Proceedings of the Human Factors and Ergonomics Society 2016 Annual Meeting 973

REFERENCES
Chaffin, D. B., Andres, R. O., & Garg, A. (1983). Volitional
postures during maximal push/pull exertions in the sagit-
tal plane. Human Factors: The Journal of the Human Fac-
tors and Ergonomics Society, 25(5), 541-550.
Charry, E., Lai, D. T., Begg, R. K., & Palaniswami, M. (2009,
January). A study on band-pass filtering for calculating
foot displacements from accelerometer and gyroscope
sensors. In Conference proceedings: Annual International
Conference of the IEEE Engineering in Medicine and Bi-
ology Society. IEEE Engineering in Medicine and Biolo-
gy Society. Conference (No. 2009, pp. 4824-4826).
Cutti, A. G., Giovanardi, A., Rocchi, L., Davalli, A., & Sac-
chetti, R. (2008). Ambulatory measurement of shoulder
and elbow kinematics through inertial and magnetic sen-
sors. Medical & biological engineering & computing,
46(2), 169-178.
da Costa, B. R., & Vieira, E. R. (2010). Risk factors for
work‐related musculoskeletal disorders: a systematic re-
view of recent longitudinal studies. American journal of
industrial medicine, 53(3), 285-323.
Duffy, V. G. (2007). A methodology for assessing industrial
workstations using optical motion capture integrated with
digital human models. Occupational Ergonomics, 7(1),
11-25.
Godwin, A., Agnew, M., & Stevenson, J. (2009). Accuracy of
inertial motion sensors in static, quasistatic, and complex
dynamic motion. Journal of biomechanical engineering,
131(11), 114501.
Granata, K. P., Lee, P. E., & Franklin, T. C. (2005). Co-
contraction recruitment and spinal load during isometric
trunk flexion and extension. Clinical Biomechanics,
20(10), 1029-1037.
Neumann, W. P., Wells, R. P., Norman, R. W., Kerr, M. S.,
Frank, J., Shannon, H. S., & Groupd, O. W. (2001). Trunk
posture: reliability, accuracy, and risk estimates for low
back pain from a video based assessment method. Interna-
tional journal of industrial ergonomics, 28(6), 355-365.
Oshima, Y., Kawaguchi, K., Tanaka, S., Ohkawara, K., Hiki-
hara, Y., Ishikawa-Takata, K., & Tabata, I. (2010). Classi-
fying household and locomotive activities using a triaxial
accelerometer. Gait & posture, 31(3), 370-374.
Ravi, N., Dandekar, N., Mysore, P., & Littman, M. L. (2005,
July). Activity recognition from accelerometer data. In
AAAI (Vol. 5, pp. 1541-1546).
Ray, S. J., & Teizer, J. (2012). Real-time construction worker
posture analysis for ergonomics training. Advanced Engi-
neering Informatics, 26(2), 439-455.
Stobbe, T. J. (1982). The development of a practical strength
testing program for industry.
Williamson, R., & Andrews, B. J. (2001). Detecting absolute
human knee angle and angular velocity using accelerome-
ters and rate gyroscopes. Medical and Biological Engi-
neering and Computing, 39(3), 294-302.
Welch, G., & Foxlin, E. (2002). Motion tracking survey. IEEE
Computer graphics and Applications, 24-38.
Wong, W. Y., & Wong, M. S. (2008). Trunk posture monitor-
ing with inertial sensors. European Spine Journal, 17(5),
743-753.
Proceedings of the Human Factors and Ergonomics Society 2016 Annual Meeting 974