A computerized video coding system for biomechanical analysis of lifting tasks
Abstract
This paper illustrates the design of a computerized postural coding system using information from field survey videotapes and limited input (e.g., load, weight, and height) to provide a timely estimation of kinematic and kinetic data for biomechanical analysis of sagittal lifting task evaluation and design. The main objectives of the study were to report the development of this technique, assess its applicability, and examine its prediction tolerance under several lifting conditions. A computer graphical user interface was developed, relying on interactive graphical three-dimensional animation to assist the prediction of a subject's lifting movement. The subject's motion was predicted based on the identification of major joint angles of key posture events extracted from the lifting video clips of the analyzed task. The key prediction outcome of this approach is the estimation of joint loading over time, including the compressive force on the lumbosacral (L5/S1) joint. Biomechanical experiments were conducted to evaluate the proposed method under several sagittal lifting conditions. The results showed that the proposed method is comparable to the use of a complex system. While there existed a tolerance between both systems in estimation of the lower back compressive force, the maximum error percentage (≈10%) is considered within a reasonable range. The implication of this approach is to provide a feasible method for performing on-site evaluations of the biomechanics of lifting tasks which cannot be performed otherwise due to the limitations of time and resources. It can also be used for pilot studies to timely identify and select the most critical tasks for more detailed analyses.
1 Figures
International Journal of Industrial Ergonomics 32 (2003) 239–250
A computerized video coding system for biomechanical
analysis of lifting tasks
Chien-Chi Chang
a,
*, Simon Hsiang
b
, Patrick G. Dempsey
a
,
Raymond W. McGorry
a
a
Liberty Mutual Research Institute for Safety, 71 Frankland Road, Hopkinton, MA 01748, USA
b
Texas Tech University, Department of Industrial Engineering, Lubbock, TX 79409, USA
Received 26 September 2002; received in revised form 6 January 2003; accepted 7 April 2003
Abstract
This paper illustrates the design of a computerized postural coding system using information from field survey
videotapes and limited input (e.g., load, weight, and height) to provide a timely estimation of kinematic and kinetic data
for biomechanical analysis of sagittal lifting task evaluation and design. The main objectives of the study were to report
the development of this technique, assess its applicability, and examine its prediction tolerance under several lifting
conditions. A computer graphical user interface was developed, relying on interactive graphical three-dimensional
animation to assist the prediction of a subject’s lifting movement. The subject’s motion was predicted based on the
identification of major joint angles of key posture events extracted from the lifting video clips of the analyzed task. The
key prediction outcome of this approach is the estimation of joint loading over time, including the compressive force on
the lumbosacral (L5/S1) joint. Biomechanical experiments were conducted to evaluate the proposed method under
several sagittal lifting conditions. The results showed that the proposed method is comparable to the use of a complex
system. While there existed a tolerance between both systems in estimation of the lower back compressive force, the
maximum error percentage (E10%) is considered within a reasonable range. The implication of this approach is to
provide a feasible method for performing on-site evaluations of the biomechanics of lifting tasks which cannot be
performed otherwise due to the limitations of time and resources. It can also be used for pilot studies to timely identify
and select the most critical tasks for more detailed analyses.
Relevance to industry
This study presents a design that is capable of performing the biomechanical assessment of manual lifting tasks using
only the field survey videotapes and limited input data. For industrial-based in situ analyses, this approach may provide
an alternative for biomechanical assessment of manual lifting tasks while still maintaining the quality of results.
r2003 Elsevier Science B.V. All rights reserved.
Keywords: Manual lifting; Biomechanical analysis; Video coding; Computer simulation
1. Introduction
Despite increased mechanization and automa-
tion of materials handling processes, manual
ARTICLE IN PRESS
*Corresponding author. Tel.: +1-508-497-0260; fax: +1-
508-435-8136.
E-mail address: chien-chi.chang@libertymutual.com
(C.-C. Chang).
0169-8141/03/$ - see front matter r2003 Elsevier Science B.V. All rights reserved.
doi:10.1016/S0169-8141(03)00065-9
materials handling (MMH) tasks are still common
in many workplaces, and are essential functions of
many jobs. Workers’ compensation losses asso-
ciated with these activities are one of the most
significant loss sources in many industry sectors
(Murphy et al., 1996).
Of the various approaches to the design and
analysis of MMH tasks, the biomechanical
approach could be used to estimate the joint
loadings applied on worker’s body in performing a
task. Many biomechanical models (e.g., Chaffin,
1969;Freivalds et al., 1984;Kromodihardjo and
Mital, 1986, 1987;Cheng and Kumar, 1991;
McGill et al., 1996) have been developed over
the years. To perform such analysis, segmental
coordination data, such as angular displacement
of major joints, are necessary component of the
analyses in most of the cases. These analyses often
require sophisticated equipment and personnel
trained to use the tools and software in order to
digitalize the movement data. Accelerometers,
goniometers, active and passive cinematographic
systems, and electromagnetic field based motion
tracking systems are all commonly used to
quantify human motion. While these methods
have the advantage of being able to accurately
record movement data, the disadvantage to all of
these systems is that in most working environ-
ments, it is difficult to duplicate the lab capacity
for capturing motion patterns in the field. Most of
these systems that are best suited for laboratory
use may not function well in the workplace.
Workplace application of biomechanical models
presents unique challenges absent in laboratory
settings.
In contrast to the needs of complex measure-
ment systems, observation measurement techni-
ques or tools (e.g., Karha et al., 1977;Corlett et al.,
1979;Chen et al., 1989;van der Beek et al., 1992;
McAtamney and Corlett, 1993) are simple to use
and may be more applicable for field studies.
These methods usually quantify and evaluate the
postures, duration, or frequency of worker per-
forming the tasks by comparing how a particular
body segment deviates from a predefined standard
position. While these approaches can easily be
implemented in the field with a single video
camera, they may be often limited to basic
information and frequently lack the detail needed
for further analysis.
The need for application of biomechanical
analyses in the workplace has led to the develop-
ment of approaches not requiring sophisticated
equipment. For example, Taha et al. (1997)
developed a neural network approach to measure
human motion without markers. This system
selects key frames from a video capture of motion
and identifies coordinate positions of the joints.
Joint angles are then calculated using these
positions based on a multiple link model of the
human body. The major drawback of this system
is that it is unable to perform real time measure-
ment and analysis and it becomes laborious as the
number of key frames increases. Chang et al.
(2000) developed an image processing-based com-
puter-aided system for quantitative analysis of
human posture and movement. Using a single
video camera, this system identifies whole-body
segments directly from the video frames. How-
ever, this system is restricted to a specific test
environment.
Computer simulation (e.g., Ayoub and El-
Bassoussi, 1978;Hsiang and Ayoub, 1994;Ber-
nard et al., 1999;Chang et al., 2001) could be one
possible option in reducing the need of data
collection for biomechanical analysis of manual
lifting tasks. However, the joint angle data
information over certain lifting periods is still
needed as input for the entire joint trajectory
prediction of manual lifting. An alternative to
direct measurement could be using the photo-
graphic method to identify and estimate the joint
posture based on the perception and visualization
of a worker’s posture during task performance. A
study of investigating the use of photographic
method to estimate the subject’s working posture
from one single photo has been reported in
previous research (Liu et al., 1997).
In this paper, we present the development of a
computerized video coding system for estimating
the loading levels experienced by the workers
during manual lifting tasks without the demands
of using complex motion tracking equipment. We
hypothesize that the individual’s joint loadings
may be approximately estimated by this proposed
design based on the posture identification from
ARTICLE IN PRESS
C.-C. Chang et al. / International Journal of Industrial Ergonomics 32 (2003) 239–250240
several photographic snapshots of the coded
manual lifting video clips and its joint trajectory
prediction using a computer simulation program.
The main purpose of the work reported here was
to introduce how this approach was designed. A
limited scope laboratory experiments were per-
formed to validate this technique and to assess its
applicability. The effects in the prediction toler-
ance under several lifting varieties (e.g., lifting
range, load, and speed) were evaluated. The results
of the biomechanical estimates of sagittal lifting
tasks performed using this system were compared
to a more traditional analysis using a motion
tracking system. The comparison would provide
insight into whether this approach is worth further
development, including its application to addi-
tional types of materials handling tasks.
2. Model development and program design
Previous research (Hsiang et al., 1998) proposed
a motion coding method to analyze a task based
on limited postural information coded from
videotapes of subjects performing manual lifting.
In the current study, the methodologies described
above were expanded and integrated with an
automated video-based postural coding program
and a joint motion prediction algorithm. A
computer graphical user interface was developed,
relying on interactive graphical three-dimensional
animation to assist the analyst. Detailed explana-
tions of how the technical bases and design
algorithm of this system were formulated are
presented in the following outlines using the
example of a computer program. The flow chart
below (Fig. 1) summarizes the steps involved in the
design of the program.
2.1. Step 1: Anthropometric data input and video
capture of lifting tasks
Anthropometric data are needed to estimate the
anthropometric parameters required for the bio-
mechanical model (e.g. body segment mass as
percentage of total body mass, location of mass
center as percentage of body segment length, and
the moment of inertia of each segment, etc.). An
interactive dialog box (Fig. 2a) prompts the user to
input the subject’s gender, height, and body
weight. The weight of the load lifted is entered
using the same dialog box.
On the right side of the screen of Fig. 2a, a
multimedia interface has been designed to capture
video taken with a camcorder or retrieve digitized
video from a storage media. The video clips are
stored using the AVI format, and time information
is stored in each picture frame. The timing
information is necessary for kinematic and kinetic
analysis. The capture rate of the video stream can
be set to up to 30 frames/s.
2.2. Step 2: Key frames selection
Once the task video is in digital format, the user
is required to select specific key picture frames by
replaying the captured video clip (Fig. 2b).
Although using a larger number of picture frames
in coding subject postures may increase the overall
accuracy, it may not be the best from a practical
viewpoint, considering the trade-off between
accuracy and the time available for coding. The
ARTICLE IN PRESS
(1)
Subject Data Input
& Video Capture
(2)
Key Lifting
Postures Selection
(3)
Joint Angles
Identification
(4)
Motion Patterns
Prediction &
Comparison
(5)
Biomechanical
Analysis Acceptable
?
Yes
No
Fig. 1. Design flow chart of the computer program.
C.-C. Chang et al. / International Journal of Industrial Ergonomics 32 (2003) 239–250 241
selection of the ‘‘minimum’’ number of key frames
and the ‘‘optimum’’ key events were based on the
suggestions of previous results (Hsiang et al.,
1998). They concluded that using four key frames
of video of a lifting task could provide sufficient
information to estimate angular joint displacement
and time information for a biomechanical analysis
of the task. These four key frames are defined by
ARTICLE IN PRESS
(a) (b)
(c) (d)
(e)
Fig. 2. Screenshots of the major steps of the video based biomechanical analysis computer program. (a) Step 1: data input and video
capture. (b) Step 2: four key lifting posture selection. (c) Step 3: major joint angle identification. (d) Step 4: predicted lifting motion
review and comparison. (e) Step 5: Results of biomechanical analysis.
C.-C. Chang et al. / International Journal of Industrial Ergonomics 32 (2003) 239–250242
the location of the external load. The four
suggested postures (key events) are: (1) initial
posture where hands grasp load, (2) frame when
load is closest to the body, (3) frame when load is
highest, and (4) final position before subject
releases the load after load placement. Instead of
coding the posture of every time frame, capturing
the posture only at the four key positions
significantly reduces the time and the memory
space needed for documentation. The assumption
of the key frame selection is based on the three
major objectives in manual lifting: (1) overcoming
the load inertia, (2) maintaining balance of upright
posture, and (3) sufficient box clearance (Hsiang
et al., 1999). Following this step, the four key
picture frames, as well as the timing data extracted
from the video file, are available to the program.
These data are used in the next two steps to
identify the subject’s joint postures and, subse-
quently, to form a trajectory prediction of the
entire lift.
2.3. Step 3: Joint angle identification
After the four key postures are selected, the user
manipulates a 3D mannequin in the interactive
window to help identify the subject’s major joint
angles (Fig. 2c) for each of the frames. Using the
3D graphical interface, the user can rotate
different camera views of the mannequin, resize
the mannequin as necessary, and adjust the joints
of the mannequin to better match the posture of
the lifter. The benefit of putting the video and the
mannequin side by side is to provide direct posture
comparison. Furthermore, the rotation of the
mannequin can reduce the need for the coder’s
mental rotation (Gaunet and Berthoz, 2000)
2.4. Step 4: Motion pattern predictions
Using the joint angle data and time information
extracted from the video described in Sections 2.2
and 2.3, a motion pattern prediction algorithm
(Hsiang et al., 1999) was adapted to generate the
entire motion pattern of the major joints. An
animation window (Fig. 2d) shows a mannequin
lifting according to the predicted trajectory. A
second window shows the synchronized video of
the subject’s lifting motion which is being simu-
lated. A scroll bar is used to control both the
animation and video replay windows. The motion
pattern can be evaluated continuously or in a
frame-by-frame comparison. The user can move
the scroll bar forward or backward repeatedly to
observe and compare the movement between the
two sets of motion patterns (subject vs. manne-
quin). Through this comparison, the user can
evaluate if the computer program has prescribed
an acceptable motion pattern. If not, the user
should go back to the previous two steps to adjust
the input before proceeding. It is obvious that the
resolution provided from a videotape with a single
viewing angle is limited. The comparison of the
motion patterns gives an opportunity to evaluate
the frame selection and postural coding based on
the human perceptual ability to recognize and
judge body movement (Chatterjee et al., 1996;
Kourtzi and Shiffrar, 1999).
2.5. Step 5: Biomechanical modeling and analysis
Based on the load, subject anthropometrics, and
kinematic data predicted by the program, kinetic
parameters such as moments applied at major
joints are derived using Newtonian dynamics
methods. Many static or dynamic biomechanical
models have been developed to estimate the load
applied on the lumbar spine joint (e.g., Chaffin,
1969;Garg et al., 1982;McGill and Norman, 1985,
1986;Freivalds et al., 1984;Lander et al., 1990;
Marras and Sommerich, 1991;Chaffin and An-
derson, 1991;McGill et al., 1996). Each model
required specific assumptions and simplifications.
Some models also considered the dynamic compo-
nents of lifting, passive tissue loading and/or
three-dimensional loading of the muscles. The
investigation of advantages and disadvantages of
each model are outside the scope of this research
and are not discussed here. In this study, the lower
back compressive forces applied at the L5/S1
joint are estimated based on an embedded
simplified biomechanical model adapted from
previous research (Lander et al., 1990;Chaffin and
Anderson, 1991).
The output results of the program include:
estimated individual anthropometric information,
ARTICLE IN PRESS
C.-C. Chang et al. / International Journal of Industrial Ergonomics 32 (2003) 239–250 243
trajectory of predicted joint angle movements, and
comparison figures between joint moments, joint
moment strengths, and L5/S1 compression forces
(see Fig. 2e).
3. Experimental investigation
An experiment was performed to investigate the
accuracy of the this simulation-based biomechani-
cal analysis relative to results from analysis using a
motion tracking system.
3.1. Subjects
Eight subjects (age: 32.379.7 yr, height:
175.9710.5 cm, weight: 89.2715.67 kg) were re-
cruited from the local area via newspaper adver-
tisement. The experimental procedure was
explained to each subject. All subjects gave written
informed consent, filled out a brief medical history
and were screened to assure they had no history of
low back pain in the past year, and no other active
musculoskeketal disease. The protocol was ap-
proved by an institutional review committee for
the protection of human subjects.
3.2. Apparatus
An adjustable workstation was set up to
simulate different lifting requirements. Subjects
were instructed to lift a container from a
predetermined initial position to a final position.
The lifting path was not constrained in any way. A
video camcorder recorded subject motion in a
plane parallel to the subject’s sagittal plane. At the
same time, an electromagnetic motion tracking
system (MotionStar Wireless, Ascension Technol-
ogy Inc., Burlington, VT, USA), was used as the
comparative measurement of joint displacement.
Fig. 3 illustrates an example of an instrumented
subject performing the manual lifting tasks in this
experimental approach.
3.3. Procedures
A total of 18 (3 32) combinations of three
lifting ranges (floor to shoulder (FS), floor to
knuckle (FK), and knuckle to shoulder (KS)),
three lifting speeds (slow (S), normal (N), and fast
(F)), and two loading conditions (9.5 and 19 kg)
were performed by each subject.
Sensors were attached to the subject’s major
joints (elbows, shoulders, hips, knees, and ankles)
and the motion tracking system was set to capture
86.1 samples per second. Joint angular kinematic
data were calculated by using joint displacement
data collected from the motion tracking system.
For all lifting trials, a biomechanical analysis was
performed to estimate the load applied on the
subject’s major joints during the tasks. These
results are based on the inputs from each subject’s
kinematic data collected by the motion tracking
system during the laboratory experiment and their
individual anthropometric data. The biomechani-
cal models used to analyze the subject’s kinematic
and kinetic information above are the same as the
models used by the computer program mentioned
in the previous section.
A coder performed the biomechanical analysis
for each lifting task based on videotapes recorded
during the laboratory experiment. All camcorder
ARTICLE IN PRESS
Fig. 3. A picture of the experimental apparatus showing one of
the instrumented subjects performing a lifting task.
C.-C. Chang et al. / International Journal of Industrial Ergonomics 32 (2003) 239–250244
videos were first digitized, separated into indivi-
dual video clips for each trial, and saved on the
computer. The coder executed the designed com-
puter program and retrieved these digitized lifting
video clips from data storage, input subject’s
information, selected the key posture frames of
the lift, identified the major joint angles of all four
key postures, and reviewed the prediction results
of lifting motion patterns. Following these steps,
the computer program generated a set of predic-
tion results for that trial.
3.4. Design and analysis
The differences between the two different
approaches were analyzed. Since peak spinal
compression is often used as a measure of
potential injury (e.g., Waters et al., 1993), this
dependent variable was used to compare the two
approaches. The results of the traditional labora-
tory approach using a motion tracking system will
be referred to as ‘‘Motion Tracking-Based’’,
whereas the results generated by the designed
computer program will be referred to as ‘‘Video-
Based’’ results. Collectively, these two methods
will be referred to as Computation Method, and
were one factor in a four-factor fully crossed
repeated measures analysis of variance (ANOVA).
The other three factors were lifting range (FS, FK,
KS), load (9.5 and 19 kg), and speed (S, N, F).
4. Results
Fig. 4 illustrates an example of L5/S1 compres-
sive forces over the entire lifting course calculated
with the two approaches.
The peak L5/S1 compressive force value was
taken from each profile and the results described
below refer to peak L5/S1 compressive forces.
Table 1 provides the mean, standard deviation,
and maximum of the absolute error (when
comparing Video-Based to Motion Tracking-
Based) for each of the experimental treatments.
The mean errors tend to be fairly small and most
of the maximum absolute errors are less than 10%.
Aside from the evaluation of overall prediction
of L5/S1 peak compressive force values shown
above, statistical analyses were also performed.
The initial ANOVA included all two- and three-
way interactions and the four-way interaction
term. An alpha level of 0.05 was used as a criterion
ARTICLE IN PRESS
0
500
1000
1500
2000
2500
3000
3500
4000
4500
0% 20% 40% 60% 80% 100%
Percentage of Total Lift Duration
L5/S1 Compressive Force (N)
Video-Based
Motion Tracking-Based
Fig. 4. An example of L5/S1 compressive force diagram (with
average absolute error=193 N and R
2
=0.76 for this contin-
uous pattern, respectively) for a single lift calculated by each of
the computation methods: Video-Based (faint curve) vs.
Motion Tracking-Based (solid curve).
Table 1
Mean, standard deviation, and maximum, of absolute error
percentages of peak L5/S1 compressive forces (Video-Based vs.
Motion Tracking-Based) for 18 lifting conditions
Load Lifting
range
Lifting
speed
Mean
error
(%)
Std.
dev. (%)
Maximum
absolute
error (%)
F 4.78 2.41 8.83
FS N 5.66 1.89 7.81
S 4.44 2.27 6.94
F 5.63 2.20 9.98
9.5 kg FK N 5.97 2.17 8.85
S 3.65 1.41 5.85
F 3.06 1.64 5.62
KS N 3.68 3.00 9.30
S 2.53 3.25 10.08
F 3.27 2.26 6.89
FS N 5.54 2.68 9.78
S 3.31 2.62 9.13
F 6.59 2.25 10.89
19 kg FK N 4.78 1.82 8.28
S 5.49 2.81 10.56
F 4.94 3.75 10.20
KS N 5.46 2.34 10.62
S 3.56 4.17 13.63
FS: floor to shoulder, FK: floor to knuckle, KS: knuckle to
shoulder, S: slow, N: normal, F: fast.
C.-C. Chang et al. / International Journal of Industrial Ergonomics 32 (2003) 239–250 245
for significance. Non-significant interactions were
removed for the sake of parsimony. Table 2
provides the ANOVA results. The table indicates
that load was the factor most influencing L5/S1
compression. All factors were significant, as were
two of the two-way interactions (Load *Range,
Computation method *Range). Figs. 5 and 6
show the interactions between load and range,
and computation method and range, respectively.
The mean L5/S1 compressive force values for
the fast lifting speed, normal lifting speed, and
slow lifting speed were significantly different from
each other. Duncan’s Multiple Range Test was
used for post hoc comparisons. Since there were
significant two-way interactions, the results for the
single factors involved in the interactions need to
be interpreted cautiously.
As Fig. 5 shows, the mean spinal compression
was higher for the 19 kg load than for the 9.5 kg
load, with a slightly higher difference for the KS
range. Overall, L5/S1 compressive force value for
the KS range was lower than for the other two
ranges.
ARTICLE IN PRESS
Table 2
ANOVA results for significant model terms
Source df Type I SS Mean square FValue pvalue
Subject 7 50,533,033.57 7,219,004.8 327.72 o0.0001
Load 1 268,04,444.74 26,804,444.74 1216.83 o0.0001
Range 2 8,833,107.1 4,416,553.35 200.5 o0.0001
Speed 2 406,132.35 203,066.18 9.22 0.0001
Computation method 1 859,368.15 859,368.15 39.01 o0.0001
Load *Range 2 578,359.9 289,179.95 13.13 o0.0001
Computation method *Range 2 428,531.14 214,265.57 9.73 o0.0001
df=degrees of freedom, SS=sum of squares.
0
1000
2000
3000
4000
5000
FS FK KS
Liftin
g
Ran
g
e
L5/S1 Compressive Forces (N)
19 kg
9.5 kg
Fig. 5. Interactive effects of loads (19 kg vs. 9.5 kg) and lifting
range (FS: floor to shoulder, FK; floor to knuckle, KS: knuckle
to shoulder) on peak L5/S1 compressive force values. The error
bars represent one standard deviation.
0
1000
2000
3000
4000
5000
FS FK KS
Liftin
g
Ran
g
e
L5/S1 Compressive Forces (N)
Video-Based
Motion Tracking-Based
Fig. 6. Interactive effects of computation method (Video-Based
vs. Motion Tracking-Based) and lifting range (FS: floor to
shoulder, FK: floor to knuckle, KS: knuckle to shoulder) on
peak L5/S1 compressive force values. The error bars represent
one standard deviation.
C.-C. Chang et al. / International Journal of Industrial Ergonomics 32 (2003) 239–250246
Fig. 6 shows that the differences between Video-
Based and Motion Tracking-Based values varied
across lifting range, with the lowest difference for
the KS range. The Video-Based values tend to be
slightly higher than Motion Tracking-Based va-
lues. Although this interaction is statistically
significant, the small differences do not appear to
have substantial practical significance, particularly
if the biomechanical results are used for relative
comparisons.
Fig. 7 illustrates the correlation between the two
methods of estimating spinal compression. There
is good correlation between the methods
(r
2
=0.93), with the Video-Based method provid-
ing slightly higher estimates.
5. Discussion
Typical field applications of biomechanical
analysis include estimating loads on the lumbar
spine and torque requirements at individual joints.
The inaccuracies inherent in various aspects of
biomechanical calculations imply that the most
appropriate use of model estimates is a relative
comparison of working conditions (Dempsey,
1998). For example, the most stressful element of
a task can be identified, or the potential reduction
in stress due to task redesign can be estimated. In
this research, we presented and examined the
feasibility of a proposed biomechanical analysis
system that alleviates the need for extensive
capture of joint kinematics. A standard videotape
is used in place of much more costly motion
tracking techniques commonly utilized in labora-
tories. The comparison results showed that the
error percentage of the overall peak L5/S1
compressive force prediction among various lifting
conditions appear to be within a tolerable range.
Given the fact that the most suitable use of
estimates of biomechanical stresses such as L5/S1
compression are relative comparisons of different
tasks (Dempsey, 1998), the prediction method
investigated could be a reasonable alternative to
ARTICLE IN PRESS
y = 0.9811x + 173.7
0
1000
2000
3000
4000
5000
0 1000 2000 3000 4000 5000
Peak L5/S1 Compressive Forces (Unit: N) - "Motion Tracking-Based"
Peak L5/S1 Compressive Forces (Unit: N) -"Video-Based"
y = x
Fig. 7. Correlation between the two methods (X-axis: Motion Tracking-Based, Y-axis: Video-Based) of estimating peak L5/S1
compressive forces (Unit: N).
C.-C. Chang et al. / International Journal of Industrial Ergonomics 32 (2003) 239–250 247
methods requiring extensive kinematic data mea-
surement using complex hardware and/or software
programs.
The system is not intended to replace ‘‘tradi-
tional’’ biomechanical task evaluations that may
provide better resolution but require complex
equipment set up and/or a laboratory-like envir-
onment. Rather it may provide an alternative
more suitable for workplace applications where
the use of a motion tracking system may be
impossible. Or it can be used for pilot evaluation
to easily identify the most critical task for more
detailed conventional analyses.
Compared to previous research (Hsiang et al.,
1998), there are several advantages to the ap-
proach described in this paper: (1) automatic
timing of lifting tasks is possible since the digitized
video clips actually store the time information in
each frame. However, there are inherent limits
depending on the PC video capture board used.
The maximum capture rate of video clips in this
system is 30 frames/s, allowing a minimum incre-
ment of 0.033 s for the four key event selections.
The use of a video capture rate limited to
30 frames/s may compromise the prediction accu-
racy of a very fast lift in comparison to the use of
traditional motion tracking systems with higher
capture rates. The lower the frame capture rate
used, the more potential there is for losing
identification of important events/information of
the lifting. (2) Enhancement of analysis. The
analysis includes joint moments and the L5/S1
compressive force over the entire lifting cycle. (3)
Error Control. There are two major sources of
errors in the joint trajectories prediction: errors in
the selection of the timing of the four key lifting
postures and errors in the identification of the
lifting postures. Although there is no way for the
user to identify the most accurate selection of
these parameters without first performing a
corresponding experiment validation, interactive
comparison of a subject’s lifting motion between
the video clip and the simulated 3D-mannequin
movement provides a method for minimizing
coding error. Both sources of errors can be
examined in real time and, at least, the prediction
of unrealistic lifting motion patterns can be
avoided. Further studies are needed to examine
in detail the effect of this approach in the
reduction of prediction error.
5.1. Limitations
There are some limitations of the approach
developed that need to be discussed. Primarily, the
approach is applicable to sagittal-plane lifting
tasks. Thus, the approach does not apply to
numerous other MMH tasks such as pushing
and pulling, or lifting tasks which involve twisting
movements (3D). In addition, only freestyle lifts
were considered and evaluated in this study. The
effects of specific lifting techniques, which subjects
could be trained or requested to perform, are not
investigated in this research. However, further
development will lead to applicability to a wider
range of tasks. Also, the biomechanical model
used in this approach does not consider muscle co-
activation that would be incorporated into elec-
tromyography-assisted models. Such in-depth
analyses are more appropriate for the laboratory,
and this is an example of what sometimes must be
sacrificed when performing field analyses.
From an operational standpoint, several factors
that might influence the accuracy of the Video-
Based approach were not studied. The data
presented were from one coder, thus future
research needs to investigate both inter- and
intra-coder reliability. Similarly, the only camera
angle used was perpendicular to the sagittal plane
of the subjects. Non-orthogonal camera angles
may induce larger errors on posture specification
(Liu et al., 1997). Additional studies will be needed
to quantify these possible influences.
6. Conclusions
In contrast to some observational task analysis
techniques, biomechanical analysis methods can
provide a more quantitative description of stresses
associated with MMH tasks. While the complex-
ities of these approaches may have limited the
application of biomechanical analysis in field
applications, the approach described here may be
a suitable alternative with less restrictive applica-
tion demands. By using a camcorder, this system is
ARTICLE IN PRESS
C.-C. Chang et al. / International Journal of Industrial Ergonomics 32 (2003) 239–250248
envisioned to provide ergonomists with a cost-
effective method for performing on-site evalua-
tions of the biomechanics of lifting tasks without
complex equipment. The usefulness of such a
system is to provide timely estimation of kinematic
and kinetic data for job design, which cannot be
performed otherwise due to the limitation of time
and resources.
The results show that, although a statistically
significant difference was found, the magnitudes of
the differences in peak L5/S1 compression force
are not of practical significance. Since the com-
pression force estimates should only be used to
compare tasks, the errors are reasonable. How-
ever, the results presented are initial and only
apply to a limited set of lifting conditions and
specific biomechanical model. Further develop-
ment is required to extend the applicability of this
approach.
References
Ayoub, M.M., El-Bassoussi, M.M., 1978. Dynamic biomecha-
nical model for sagittal plane lifting activities. Safety in
Manual Materials Handling, DHEW/NIOSH Publication
No. 78-185.
Bernard, T.M., Ayoub, M.M., Lin, C.J., 1999. Evaluation of a
biomechanical simulation model for sagittal plane lifting.
International Journal of Industrial Ergonomics 24, 157–171.
Chaffin, D.B., 1969. A computerized biomechanical model—
development of and use in studying gross body actions.
Journal of Biomechanics 2, 429–441.
Chaffin, D.B., Anderson, C.K., 1991. Occupational Biomecha-
nics. Wiley, New York.
Chang, R., Guan, L., Burne, J.A., 2000. An automated form of
video image analysis applied to classification of movement
disorders. Disability and Rehabilitation 22 (1/2), 97–108.
Chang, C.C., Brown, D.R., Bloswick, D.S., Hsiang, S.M., 2001.
Biomechanical simulation of manual lifting using spacetime
optimization. Journal of Biomechanics 34, 527–532.
Chatterjee, S.H., Freyd, J.J., Shiffrar, M., 1996. Configural
processing in the perception of apparent biological motion.
Journal of Experimental Psychology: Human Perception
and Performance 22 (4), 916–929.
Chen, J.G., Peacock, J.B., Schlegel, R.E., 1989. An observation
technique for physical work stress analysis. International
Journal of Industrial Ergonomics 3, 167–176.
Cheng, C., Kumar, S., 1991. A three-dimensional static torso
model for the six human lumbar joints. International
Journal of Industrial Ergonomics 7, 327–339.
Corlett, E.N., Madeley, S.J., Maneneica, I., 1979. Posture
targeting: a technique for recording working postures.
Ergonomics 22, 357–366.
Dempsey, P.G., 1998. A critical review of biomechanical,
epidemiological, physiological and psychophysical criteria
for designing manual materials handling tasks. Ergonomics
41 (1), 73–88.
Freivalds, A., Chaffin, D.B., Garg, A., Lee, K.S., 1984.
A dynamic biomechanical evaluation of lifting
maximum acceptable loads. Journal of Biomechanics 17
(4), 251–262.
Garg, A., Chaffin, D.B., Freivalds, A., 1982. Biomechanical
stresses from manual load lifting: a static vs. dynamic
evaluation. Transactions of the American Institute of
Industrial Engineers 14, 272–281.
Gaunet, F., Berthoz, A., 2000. Mental rotation for spatial
environment recognition. Brain Research. Cognitive Brain
Research 9 (1), 91–102.
Hsiang, S.M., Ayoub, M.M., 1994. Development of methodol-
ogy in biomechanical simulation of manual lifting. Interna-
tional Journal of Industrial Ergonomics 13, 271–288.
Hsiang, S.M., Brogmus, G.E., Martin, S.E., Bezverkhny, I.B.,
1998. Video based lifting technique coding system. Ergo-
nomics 41 (3), 239–256.
Hsiang, S.M., Chang, C.C., McGorry, R.W., 1999.
Development of a set of equations described joint trajec-
tories during para-sagittal lifting. Journal of Biomechanics
32, 871–876.
Karha, O., Jansi, P., Kuorinka, I., 1977. Correcting working
postures in industry: a practical method for analysis.
Applied Ergonomics 8, 199–201.
Kourtzi, Z., Shiffrar, M., 1999. Dynamic representations of
human body movement. Perception 28, 49–62.
Kromodihardjo, S., Mital, A., 1986. Kinetic analysis of manual
lifting activities: Part I—development of a three dimensional
computer model. International Journal of Industrial Ergo-
nomics 1, 77–90.
Kromodihardjo, S., Mital, A., 1987. Biomechanical analysis of
manual lifting tasks. Journal of Biomechanical Engineering
109, 132–138.
Lander, J.E., Simonton, R.L., Giacobbe, J.K.F., 1990. The
effectiveness of weight-belts during the squat exercise.
Medicine and Science in Sports and Exercise 22 (1),
117–126.
Liu, Y., Zhang, X., Chaffin, D., 1997. Perception
and visualization of human posture information for
computer-aided ergonomic analysis. Ergonomics 40 (8),
818–833.
Marras, W., Sommerich, C., 1991. A three dimensional motion
model of loads on the lumbar spine, I: model structure.
Human Factors 33, 123–137.
McAtamney, L., Corlett, E.N., 1993. RULA: a survey method
for the investigation of work-related upper limb disorders.
Applied Ergonomics 24 (2), 91–99.
McGill, S.M., Norman, R.W., 1985. Dynamically and statically
determined low back movements during lifting. Journal of
Biomechanics 18, 877–885.
ARTICLE IN PRESS
C.-C. Chang et al. / International Journal of Industrial Ergonomics 32 (2003) 239–250 249
McGill, S.M., Norman, R.W., 1986. Partitioning of the 14–15
dynamic moment into disc, ligamentous, and muscular
components during lifting: a verification study. Spine 11,
666–678.
McGill, S.M., Norman, R.W., Cholewicki, J., 1996. A simple
polynomial that predicts low-back compression during
complex 3D-tasks. Ergonomics 39 (9), 1107–1118.
Murphy, P.L., Sorock, G., Courtney, T.K., Webster, B.S.,
Leamon, T.L., 1996. Injury and illness in the American
workplace: a comparison of data sources. American Journal
of Industrial Medicine 30, 130–141.
Taha, Z., Brown, R., Wright, D., 1997. A neural network
approach to markerless measurement of motion. Biomedical
Sciences Instrumentation 33, 441–446.
van der Beek, A.J., Gaalen, A.C., Frings-Dresen,
M.H.W., 1992. Working postures and activities of lorry
drivers: a reliability study of on-site observation and
recording on a pocket computer. Applied Ergonomics 23,
331–336.
Waters, T.R., Putz-Anderson, V., Garg, A., Fine, L.J., 1993.
Revised NIOSH equation for the design and evaluation of
manual lifting tasks. Ergonomics 36 (7), 749–776.
ARTICLE IN PRESS
C.-C. Chang et al. / International Journal of Industrial Ergonomics 32 (2003) 239–250250
- CitationsCitations25
- ReferencesReferences36
- However , in accordance with abovementioned trade-off, such methods are often costly and difficult to apply to a field setting, as they may interfere with the work performed (Trask et al., 2007). Posture fitting on planar video recordings during manual materials handling tasks has been shown to be a feasible and accurate method for application in field settings (Chang et al., 2003; Coenen et al., 2011; Xu et al., 2011). Yet, such methods are time-consuming and only allow analysis of selected tasks rather than continuous monitoring.
[Show abstract] [Hide abstract] ABSTRACT: The trade-off between feasibility and accuracy of measurements of physical exposure at the workplace has often been discussed, but is unsufficiently understood. We therefore explored the effect of two low-back loading measurement tools with different accuracies on exposure estimates and their associations with low-back pain (LBP). Low-back moments of 93 workers were obtained using two methods: a moderately accurate observation-based method and a relatively more accurate video-analysis method. Group-based exposure metrics were assigned to a total of 1131 workers who reported on their LBP status during three follow-up years. The two methods were compared regarding individual and group-based moments and their predictive value for LBP. Differences between the two methods for peak moments were high at the individual level and remained substantial at group level. For cumulative moments, differences between the two methods were attenuated as random inaccuracies cancelled out. Peak moments were not predictive for LBP in any method while cumulative moments were, suggesting comparable predictive values of the two methods. While assessment of low-back load improves from investing in collecting relatively more accurate individual-based data, this does not necessarily lead to better predictive values on a group level, especially not for cumulative loads. Copyright © 2015 Elsevier Ltd and The Ergonomics Society. All rights reserved.- Task analysis assisted the analysis of the job by dividing the main task into its components to define noticeably cognitive and physical tasks, it provided a better comprehension of the interaction and identified those subtasks that would require training like inspection and verification of data. Combining REBA and biomechanical analysis was considered effective and useful since postural methods like REBA or RULA base the evaluation from snapshots of postures, and biomechanical methods can improve the analysis by quantifying more objectively [47, 48, 54, 55] the risk for MSDs. This study presents a novel problematic since we found a lack of literature about ergonomic problems or studies during repairing computer tasks.
[Show abstract] [Hide abstract] ABSTRACT: Background: Manufacturing industry workers who repair computers may be exposed to ergonomic risk factors. Objectives: This project analyzes the tasks involved in the computer repair process to (1) find the risk level for musculoskeletal disorders (MSDs) and (2) propose ergonomic interventions to address any ergonomic issues. Methods: Work procedures and main body postures were video recorded and analyzed using task analysis, the Rapid Entire Body Assessment (REBA) postural method, and biomechanical analysis. Results: High risk for MSDs was found on every subtask using REBA. Although biomechanical analysis found an acceptable mass center displacement during tasks, a hazardous level of compression on the lower back during computer's transportation was detected. Conclusions: This assessment found ergonomic risks mainly in the trunk, arm/forearm, and legs; the neck and hand/wrist were also compromised. Opportunities for ergonomic analyses and interventions in the design and execution of computer repair tasks are discussed.- The workstation analysis function consists of three components: a work-recorded video playback, the workstation analysis table and an analysis result presentation. This function allows a user to analyse work processes in a workstation from various perspectives repeatedly, while watching a work-recorded video (Chang et al. 2003; Forsman et al. 2002; Hanse and Forsman 2001). The user then records the analysis results on the table and the recorded results are visualised.
[Show abstract] [Hide abstract] ABSTRACT: This paper proposes a framework for the continuous performance improvement of manned assembly lines for major appliances. The suggested framework consists of two main activities: (1) assembly work process improvement, including time and motion study and (2) improvement of line balance efficiency. Although there have been numerous studies on this topic, most of them deal with partial issues rather than the continuous performance improvement of the whole assembly line, which this paper addresses. To develop the framework, we categorised a manned assembly line into five analysis levels, (workstation, worker, operation cycle, work element and unit motion), and identified user requirements on each level. Among these five analysis levels, the workstation level supports line balancing, while the other four (worker, operation cycle, work element and unit motion) address work process improvement. The proposed framework has been implemented and tested with various examples from Korean assembly line based manufacturers.- (1) still holds. The L5/S1 moment can then be estimated by the kinematics of each body segment based on the identified joint angular trajectories and anthropometry data (Chang et al., 2003; Hsiang et al., 1998). However, when foot movement or the rotation of the sagittal plane exists, such a bodycentered reference frame will be a non-inertial reference frame and the effect of fictitious force needs to be considered when calculating L5/S1 moments (Jha, 2005) In a non-inertial reference frame, the fictitious force on a point mass is written as (Jha, 2005):
[Show abstract] [Hide abstract] ABSTRACT: In ergonomics studies, linked segment models are commonly used for estimating dynamic L5/S1 joint moments during lifting tasks. The kinematics data input to these models are with respect to an arbitrary stationary reference frame. However, a body-centered reference frame, which is defined using the position and the orientation of human body segments, is sometimes used to conveniently identify the location of the load relative to the body. When a body-centered reference frame is moving with the body, it is a non-inertial reference frame and fictitious force exists. Directly applying a linked segment model to the kinematics data with respect to a body-centered non-inertial reference frame will ignore the effect of this fictitious force and introduce errors during L5/S1 moment estimation. In the current study, various lifting tasks were performed in the laboratory environment. The L5/S1 joint moments during the lifting tasks were calculated by a linked segment model with respect to a stationary reference frame and to a body-centered non-inertial reference frame. The results indicate that applying a linked segment model with respect to a body-centered non-inertial reference frame will result in overestimating the peak L5/S1 joint moments of the coronal plane, sagittal plane, and transverse plane during lifting tasks by 78%, 2%, and 59% on average, respectively. The instant when the peak moment occurred was delayed by 0.13, 0.03, and 0.09s on average, correspondingly for the three planes. The root-mean-square errors of the L5/S1 joint moment for the three planes are 21Nm, 19Nm, and 9Nm, correspondingly.- The existing approaches [3], [4] require sophisticated instrumentation which can only be achieved in a laboratory environment, therefore limiting their application in industrial settings. To address this limitation, video-based biomechanics tools were developed [5]. These new tools use single camera to acquire the whole body movement and joint locations.
[Show abstract] [Hide abstract] ABSTRACT: We propose to use the Microsoft Kinect depth sensor for assessing spinal loading, and present an initial study on a public action database with 20 actions. It is built on the recent success on human skeleton extraction with depth sensor, which has been coded into the Kinect software development kit. This letter is an essential step toward a systematic evaluation in future.- However, this study did not report the error of peak 3-D back load which, according to McGill (1997), was believed to play an important role on the back injury. The 2-D dynamic model used in the video coding method discussed in the previous paragraph (Hsiang et al., 1998; Chang et al., 2003 Chang et al., , 2010) was unable to investigate the peak 3-D L5/S1 joint loading estimation for asymmetric lifting tasks. This was because, according to Kingma et al. (1998a), applying a 2-D biomechanical model to asymmetric lifting tasks would introduce error on the anterioreposterior position of the L5/S1 joint during trunk rotation.
[Show abstract] [Hide abstract] ABSTRACT: The aim of the study was to investigate the validity and interrater reliability of using a proposed video coding system to estimate the dynamical 3-D L5/ S1 joint moment on the basis of four key frames from video clips of asymmetric lifting tasks. L5/S1 joint loading has been widely adopted to quantify low-back loading during lifting tasks. However, the measurement of L5/S1 joint loading usually requires a laboratory environment, which cannot be applied during field surveys. The validity of this system was investigated by comparing the estimated L5/S1 joint moments of various simulated lifting tasks with motion tracking system-based reference L5/S1 joint moments. The comparison showed that the video coding system yielded good estimates on peak moment (r = .91, average absolute error [AAE] = 20.3 Nm) and cumulative moment (r = .88,AAE = 22.5 Nm.sec) of the sagittal plane. The interrater reliability of this system was assessed among 10 raters who used this system. The intraclass correlation ranged between .51 and .89 for the moments of different planes. The results of the validity and interrater reliability analyses showed that the proposed video coding system could provide a good estimate of total L5/S1 joint loading on the basis of side-view video clips of the simulated lifting tasks. Although it was not as accurate as a motion tracking system for L5/S1 joint loading calculations, this approach can be an alternative for back load estimation for some lifting configurations when the use of motion tracking systems is not possible.
Data provided are for informational purposes only. Although carefully collected, accuracy cannot be guaranteed. Publisher conditions are provided by RoMEO. Differing provisions from the publisher's actual policy or licence agreement may be applicable.
This publication is from a journal that may support self archiving.
Learn more