Motion Data-Driven Biomechanical Analysis during

Construction Tasks on Sites

JoonOh Seo 1; Richmond Starbuck 2; SangUk Han, A.M.ASCE 3;
SangHyun Lee, A.M.ASCE 4; and Thomas J. Armstrong 5
Downloaded from ascelibrary.org by Hong Kong Polytechnic University on 09/15/23. Copyright ASCE. For personal use only; all rights reserved.

Abstract: Work-related musculoskeletal disorders (WMSDs) are one of the major health issues that workers frequently experience due to
awkward postures or forceful exertions during construction tasks. Among available job analysis methods, biomechanical models have been
widely applied to assess musculoskeletal risks that may contribute to the development of WMSDs based on motion data during occupational
tasks. Recently, with the advent of vision-based motion capture approaches, it has become possible to collect the motion data required for
biomechanical analysis under real conditions. However, vision-based motion capture approaches have not been applied to biomechanical
analysis because of compatibility issues in body models of the motion data and computerized biomechanical analysis tools. To address this
issue, automated data processing is focused on to convert motion data into available data in existing biomechanical analysis tools, given the
BVH motion data from vision-based approaches. To examine the feasibility of the proposed motion data processing, an experiment for both
static and dynamic biomechanical analyses was conducted on lifting tasks. The results indicate that vision-based motion capture data—
converted as proposed in this paper—can provide a sufficient level of detail on human kinematics to conduct biomechanical analysis, thus
allowing for the identification of particular body parts where excessive forces are placed during tasks. The issues and directions of future
research are also discussed to perform on-site biomechanical analysis during construction tasks. DOI: 10.1061/(ASCE)CP.1943-5487
.0000400. © 2014 American Society of Civil Engineers.
Author keywords: Work-related musculoskeletal disorders; Biomechanical analysis; Motion data.

Introduction (Meerding et al. 2005), and temporary or permanent disability by

WMSDs have an adverse effect on a worker’s livelihood and
The construction industry is labor intensive and relies largely on self-esteem (Abásolo et al. 2012). In addition, great economic burden
manual handling tasks. Because of physically demanding construc- by WMSDs rests on construction firms due to lost productivity
tion tasks involving forceful exertions with awkward postures, and increased workers’ compensation costs (Albers and Estill 2007).
workers frequently suffer from work-related musculoskeletal disor- Biomechanical models—one of the ergonomic job analysis
ders (WMSDs) such as sprains, tendonitis, carpal tunnel syndrome, methods—have been widely applied to assess musculoskeletal
and back pain (Boschman et al. 2012; CPWR 2013). Statistics risks that may contribute to the development of WMSDs during
show that WMSDs accounted for 24% of nonfatal occupational occupational tasks by estimating internal loads as a function of ex-
injuries and illnesses in U.S. construction workers in 2011 (BLS ternal exposure data (e.g., postures, movements, force exertions on
2012), which means that construction workers are at about a 50% hands and feet; Chaffin et al. 2006). Because an estimation of in-
higher risk of WMSDs than workers in other industries (Schneider ternal loads requires tedious computations with three-dimensional
2001). These WMSDs are a leading cause of lost work days (3D) whole-body biomechanical models, several computerized
software packages such as three-dimensional static strength predic-
1
Ph.D. Student, Tishman Construction Management Program, Dept. of tion program (3D SSPP), OpenSim, Visual 3-D, and AnyBody have
Civil and Environmental Engineering, Univ. of Michigan, 2350 Hayward provided practical solutions to study musculoskeletal stresses.
St., Ann Arbor, MI 48109. E-mail: [email protected] Even though these computerized tools have provided compre-
2
Undergraduate Research Assistant, Tishman Construction Manage- hensive physical stress analysis for diverse occupational tasks, the
ment Program, Dept. of Civil and Environmental Engineering, Univ. of
use of these tools has been applied in only limited or controlled
Michigan, 2350 Hayward St., Ann Arbor, MI 48109. E-mail: rstarbuc@
umich.edu
environments due to the difficulty of collecting and analyzing
3
Assistant Professor, Dept. of Civil and Environmental Engineering, the motion data required for biomechanical models. However, com-
Univ. of Alberta, 9105 116th St., Edmonton, AB, Canada T6G 2W2. pared with other industries such as manufacturing where work
E-mail: [email protected] methods and processes are usually fixed when designing workpla-
4
Assistant Professor, Tishman Construction Management Program, ces, construction takes place in unstructured and varying environ-
Dept. of Civil and Environmental Engineering, Univ. of Michigan, 2350 ments, and thus work methods and postures are changing over time.
Hayward St., Ann Arbor, MI 48109 (corresponding author). E-mail: Further, task requirements in construction vary depending on
[email protected] project-specific context (Mitropoulos and Memarian 2013), which
5
Professor, Dept. of Industrial and Operations Engineering, Univ. of
results in different levels of physical exertion by workers’ muscu-
Michigan, 2350 Hayward St., Ann Arbor, MI 48109. E-mail: [email protected]
Note. This manuscript was submitted on November 3, 2013; approved
loskeletal systems. As a result, an effective and easily accessible
on March 25, 2014; published online on July 11, 2014. Discussion period means for on-site biomechanical analysis is required to assess risk
open until December 11, 2014; separate discussions must be submitted for factors (e.g., awkward postures, forceful exertions) that may cause
individual papers. This paper is part of the Journal of Computing in Civil excessive musculoskeletal stresses beyond human capability during
Engineering, © ASCE, ISSN 0887-3801/B4014005(13)/$25.00. construction tasks under real conditions.

© ASCE B4014005-1 J. Comput. Civ. Eng.

J. Comput. Civ. Eng., 2015, 29(4): B4014005

 The implementation of on-site biomechanical analysis may of WMSDs is that the risk of WMSDs increases when a worker is
broadly involve two technical challenges. One is to collect accurate excessively exposed to risk factors beyond the worker’s functional
motion data without interfering with ongoing works in construction capacity (Kumar 1999). Radwin et al. (2001) recognized the
sites; the other is to process the motion data to make it compatible interrelatedness of risk factors causing WMSDs, and identified that
with existing computerized biomechanical analysis tools. With re- external loads produced in the physical work environment are
gard to the former, previous studies have proposed vision-based transmitted to the body through a biomechanical pathway.
motion capture using ordinary video or network surveillance cam- Because WMSDs develop over time due to repeated exposure to
eras (Han et al. 2012b, 2013a) and an RGB-D sensor (e.g., Kinect; physical, biomechanical, and physiological risk factors, corrective
Han et al. 2012a, 2013b), which allows to collect motion data with- actions by identifying and eliminating any risks of WMSDs should
out invasive measures under real conditions. However, the human be taken before the symptoms get worse (Simoneau et al. 1996).
skeleton-based motion data extracted from motion capture systems Previously, a wide range of ergonomic methods such as self-reports
are not readily applicable to existing biomechanical analysis tools. (e.g., interviews and questionnaires), observational methods, and
Downloaded from ascelibrary.org by Hong Kong Polytechnic University on 09/15/23. Copyright ASCE. For personal use only; all rights reserved.

For example, 3D SSPP (Chaffin et al. 2006)—the only available direct measurements (e.g., sensors) have been developed to assess
computerized static biomechanical analysis tool designed to risk factors of WMSDs during occupational tasks (Li and Buckle
evaluate static postures or motions without acceleration—defines a 1999; David 2005). Inyang et al. (2012) emphasized on an appli-
human posture with horizontal and vertical angles in a global co- cation of these ergonomic assessments during the planning and
ordinate system (i.e., defined for a full body). OpenSim (Delp et al. execution of construction tasks not only to improve safety issues,
2007)—one of the popular dynamic analysis tools that considers but also to reduce the costs associated with WMSDs. In construc-
inertial forces caused by acceleration—simulates motions using tion, Han et al. (2013b) applied observational methods based on
marker-based motion data that contains positions of markers rather vision-based motion capture to detect unsafe actions during con-
than body joints (i.e., more than one marker generally is attached to struction tasks. Ray and Teizer (2012) also suggested the automatic
one body joint). However, vision-based motion capture systems ergonomic monitoring system to identify nonergonomic postures
generally characterize motions using Euler rotation angles at a body using RGB-D sensors. Dai and Ning (2013) pointed out that remote
joint in a local coordinate system (i.e., defined for each body joint), sensing technologies using surveillance cameras have great poten-
for example, as in the biovision hierarchical (BVH) format tial as a video-based assessment method for identifying the risk of
(Meredith and Maddock 2001). In this regard, a motion data recon- WMSDs. In addition, sensors such as physical status monitors
figuration that converts vision-based motion capture data into the (i.e., heart rate sensor; Gatti et al. 2010) or anisotropic magneto-
proper form for ergonomic analysis tools is the key to the success- resistive (AMR) sensors (i.e., body angle measurement system;
ful implementation of on-site biomechanical analysis. Alwasel et al. 2012) attached to workers have been suggested as
This paper thus proposes on-site biomechanical analysis as a a way to directly measure the physical or postural risk factors
way to assess the risk of WMSDs during construction tasks using of WMSDs during construction tasks.
vision-based motion data. Specifically, automated data processing Although these methods focused on the identification of exter-
was conducted to convert motion data into available data in existing nal risk factors, biomechanical models have helped to understand
biomechanical analysis tools, given the BVH motion data from how external factors create musculoskeletal stresses such as joint
vision-based approaches. As biomechanical analysis tools to be moments or muscle forces that can rarely be measured directly
studied, 3D SSPP and OpenSim are selected because they are (Chaffin et al. 2006). Based on the assumption that the actions
not only widely used static and dynamic biomechanical analysis of the human body follow the laws of Newtonian mechanics, the
tools, but they also enable us to customize existing functionality biomechanical models provide a quantitative assessment of the
to run biomechanical analysis. Then, an experiment for both static musculoskeletal loads during occupational tasks, given kinematic
and dynamic biomechanical analyses was conducted on lifting tasks information such as postures and motions during occupational tasks
with postural variations not only to examine the feasibility of the (Chaffin et al. 2006). As a result, biomechanical models help one to
proposed motion data processing, but also to understand the need identify hazardous loading conditions that exceed human’s physical
for on-site biomechanical analysis. Based on the results, the possibil- capability (Marras and Radwin 2005). However, it is very difficult to
ity of on-site biomechanical analysis as a field-based ergonomic measure postures and motions because of the numerous articulations
evaluation method is discussed by addressing how motions during of the human body and the multiple degrees of freedom at body
construction tasks can be understood from the biomechanical analy- joints (Monroe Keyserling and Budnick 1987). To collect motion
sis results to identify the risk of WMSDs. The contributions and data, previous biomechanical studies have relied on complex motion
potential issues when applying motion data from vision-based ap- measurement systems such as video-based posture analysis, optical
proaches for on-site biomechanical analysis were also discussed. motion capture systems, or body-fixed sensors.
In conclusion, the findings of this paper are summarized. Video-based motion analysis systems rely on two-dimensional
(2D) photographic images to measure postural angles of the
human body through visual estimation, mathematical calculations
Biomechanical Analysis for Risk Identification of (Paul and Douwes 1993), or a flexible 3D mannequin (Monroe
WMSDS Keyserling and Budnick 1987). However, even though these meth-
ods are cost-effective and noninvasive, they have limitations; they
Previous research efforts have identified mechanisms for WMSDs are time-consuming, and large errors still exist (Chaffin et al.
to describe various factors that may play into the development 2006). Optical motion capture systems using passive (e.g., retrore-
of musculoskeletal disorders, and to address biomechanical and flective) or active markers (e.g., light-emitting diode) attached to
physiological aspects and interactions between those factors a subject’s body parts are the most common techniques for collect-
(Kumar 1999). The factors are categorized into external factors ing kinematic data for biomechanical studies (Davis et al. 1991;
(e.g., awkward postures and forceful exertions) and internal factors Aminian and Najafi 2004). Recently, various commercially avail-
referring to musculoskeletal stresses (e.g., joint moments and able systems—such as VICON, Qualysis, Optotrak, and others—
muscle forces) and metabolic demands (e.g., muscle fatigue; have provided whole-body motion data with a positional accuracy
Armstrong et al. 1996). The underlying hypothesis of mechanisms of less than 1 mm (Wiles et al. 2004). Because optical motion

© ASCE B4014005-2 J. Comput. Civ. Eng.

J. Comput. Civ. Eng., 2015, 29(4): B4014005

 capture systems require a large, dedicated, controlled environment, extracted 3D skeleton models consisting of 14 key body joints
previous biomechanical studies on construction tasks generally (e.g., head, neck, shoulders, elbows, wrists, knees, and ankles) us-
collected motion data by mimicking the tasks through laboratory ing ordinary video or network surveillance cameras. These skeleton
experiments (Jia et al. 2011; Kim et al. 2011). In contrast with models can be converted into the BVH motion data format (Chun
optical motion capture systems, body-fixed sensors—such as a et al. 2003) that is a standard representation of movements based on
goniometer, an accelerometer, and a gyroscope—have flexibility in skeleton models. In addition, an RGB-D sensor-based motion cap-
use, which enables us to collect kinematic information anywhere ture approach (e.g., Kinect) provides motion data in the BVH for-
(Morris 1973; Aminian and Najafi 2004). Recently, inertial meas- mat. However, despite their potential as effective noninvasive
urement units (IMUs)-based motion capture systems (e.g., Xsens) measures of motions under real conditions, vision-based motion
using a combination of accelerometers and gyroscopes—and some- capture approaches have not been applied to biomechanical analy-
times also magnetometers—provide a more plausible solution for sis because of compatibility issues in body models of the motion
body motion capture by complementing the drawbacks of different data and computerized biomechanical analysis tools.
Downloaded from ascelibrary.org by Hong Kong Polytechnic University on 09/15/23. Copyright ASCE. For personal use only; all rights reserved.

sensors. However, for whole-body motion capture, the body-fixed

sensors should be attached to all body segments, and thus they may
interfere with ongoing works when collecting motion data under Motion Data-Driven Biomechanical Analysis
real conditions.
Recently, vision-based motion capture approaches have been This research is undertaken to explore the usability of vision-based
considered attractive solutions to the limitations (e.g., need for con- motion capture data for on-site biomechanical analysis that helps
trolled environments and a possibility of interfering with ongoing assess workers’ postures or movements during construction tasks.
work) on existing motion capture systems for biomechanics and As biomechanical analysis tools to be studied, 3D SSPP and Open-
clinical problems (Corazza et al. 2006; Moeslund et al. 2006). Vast Sim are selected because they can be partially or fully customized
research efforts have developed emerging computer vision tech- depending on the user’s needs. For example, 3D SSPP enables the
niques or algorithms to extract motion data from video cameras automatic analysis of tasks using a batch file that follows a specific
(e.g., 2D images; Moeslund and Granum 2001; Moeslund et al. format with all of the information including postural angles re-
2006; Poppe 2007). In construction, Han et al. (2012b, 2013a) quired for biomechanical analysis (Center for Ergonomics, Univer-
applied a multiple camera–based approach to extract motion data sity of Michigan 2011). OpenSim is an open-source platform
during ladder climbing from different viewpoints of images using written in C++, and provides an application programming interface
2D pose estimation and 3D reconstruction algorithms. In addition, (API) that allows researchers to access and customize OpenSim
the use of RGB-D images (i.e., 2D images plus depth information) functionality (Anderson et al. 2012). This chapter provides the de-
collected from RGB-D sensors (e.g., Kinect) has simplified the tails on the suggested processes that automatically convert the BVH
process for vision-based motion capture algorithms (Shotton et al. motion data from vision-based motion capture approaches into
2013). Several computer vision algorithms have been developed to available file formats in existing biomechanical analysis tools,
estimate human poses by detecting the 3D positions of body joints 3D SSPP and OpenSim, thus allowing us to perform biomechanical
directly from RGB-D images (Lee and Cohen 2006; Plagemann analysis using the motion data without any time-consuming data
et al. 2010; Siddiqui and Medioni 2010; Shotton et al. 2013). processing.
Recently, motion capture solutions such as iPi Desktop Motion
Capture (www.ipisoft.com) and OpenNI (http://www.openni.org)
have provided effective solutions for extracting skeleton-based mo- Automated Motion Data Processing for Static
Biomechanical Analysis in 3D SSPP
tion data from RGB-D images obtained by RGB-D sensors. The
main advantage of the vision-based motion capture is that it does 3D SSPP is static biomechanical analysis software developed by
not require any markers or sensors attached to the subject during the Center of Ergonomics at the University of Michigan (Chaffin
motion capture, and allows researchers to generate human skeleton- et al. 2006). With posture data, anthropometry data, and force
based motion data (Fig. 1 shows examples of human skeleton- parameters, workers’ motions can be simulated in a virtual 3D envi-
based motion data). For example, Han et al. (2012b, 2013a) ronment. Based on the biomechanical simulation, static strength

Fig. 1. Skeleton-based motion data: (a) 3D skeleton; (b) an example of skeleton model in BVH motion data

© ASCE B4014005-3 J. Comput. Civ. Eng.

J. Comput. Civ. Eng., 2015, 29(4): B4014005

Downloaded from ascelibrary.org by Hong Kong Polytechnic University on 09/15/23. Copyright ASCE. For personal use only; all rights reserved.

Fig. 2. 3D SSPP: (a) user interface; (b) angular configurations of a human model

requirements (e.g., joint moments) for certain tasks are predicted, epidemiologically and biomechanically (Jäger and Luttmann
including the spinal compression force using the static biomechani- 1999). The joint moments below the AL can be achieved by
cal model (Center for Ergonomics, University of Michigan 2011) 99% of men and 75% of women, which means almost every type
that assumes the effects of acceleration and momentum are negli- of worker can perform the task. On the other hand, the joint mo-
gible. More important, based on the analysis results of postures, the ments beyond the MPL can be exerted only by 25% of men and 1%
body parts that endure forceful exertion can be found as compared of women, and thus should not be permitted to prevent musculo-
with the relevant human capacity such as the National Institute for skeletal injuries. For back compression forces, the AL is set to
Occupational Safety and Health (NIOSH)-recommended limits for 3,400 N, and the MPL is set to 6,400 N. As a result, if the bar that
percent capables (i.e., percent of the population with sufficient represents joint moments and back compression forces in certain
strength; Center for Ergonomics, University of Michigan 2011). postures is in the red zone, the body segments have a high risk
For example, Fig. 2(a) shows an example of the biomechanical of getting injured.
analysis result in 3D SSPP. The left three images in Fig. 2(a) Generally, a static biomechanical model requires three types of
are the same pose from different viewpoints, and the right image input data: (1) anthropometric factors (body lengths, masses, and
in Fig. 2(a) shows the analysis result. The limits in the bar centers of mass of body segments), (2) force parameters (external
graphs correspond to the NIOSH action limit (AL) and maximum forces exerted on hands and feet), and (3) body angles at each body
permissible limit (MPL; NIOSH 1981) that were substantiated joint. In 3D SSPP, anthropometric factors are set as default values

Fig. 3. Work flow for automated motion data processing in 3D SSPP

© ASCE B4014005-4 J. Comput. Civ. Eng.

J. Comput. Civ. Eng., 2015, 29(4): B4014005

 for a U.S. industrial population and can be adjusted based on the The postural angles calculated from the BVH motion data for
subject’s height and weight. Force parameters referring to hand each frame are integrated in a batch file that allow automatic analy-
forces during lifting, pushing, and pulling—foot forces are deter- sis of tasks just by importing the batch file into 3D SSPP (Center
mined by summing body weight and hand force vectors—specify for Ergonomics, University of Michigan 2011). Fig. 4(a) shows
external forces during tasks. While anthropometric factors and an example of a batch file automatically generated, containing in-
force parameters can be included by simply inputting the subject’s formation to run a biomechanical analysis in 3D SSPP. All lines in
height and weight, and hand forces in 3D SSPP, postural angles the batch file have one command describing relevant data. Types
should be determined from motion data. The body model in 3D and functions of commands used in a batch file are illustrated in
SSPP defines a posture as body segment angles with three degrees Fig. 4(b). For example, ANTHROPOMETRY, HANDLOADS,
of freedom, and thus can be manipulated by inputting the angles for and SEGMENTANGLES commands are for inputting anthropom-
each frame, as shown in Fig. 2(b). etry data (gender, height, and weight), hand forces required to per-
This paper proposed an automated process to convert the BVH form the tasks, and body angles. Specifically, the values for body
Downloaded from ascelibrary.org by Hong Kong Polytechnic University on 09/15/23. Copyright ASCE. For personal use only; all rights reserved.

motion data into postural angles defined in 3D SSPP, and then to angles are computed directly from the BVH motion data. The other
run biomechanical analysis, given the BVH motion data as shown commands—such as COMMENT, DESCRIPTION, AUTOEX-
in Fig. 3. BVH motion data defines hierarchical body segments as PORT, FRAME, and EXPORT—are used to set configurations
local translation and rotation information from a root body joint of output data (.exp). By running this batch file in 3D SSPP, an
(e.g., a hip). However, the definitions of rotation angles and the external text file (.exp) containing the results (e.g., summary re-
coordinate system in BVH motion data are different from the def- sults, joint moments, back compression forces, and strength capa-
initions and coordinate system in 3D SSPP. To address the differ- bilities) from a biomechanical analysis can be generated.
ence, the body angles required for 3D SSPP were computed based
on spatial information (local translations and rotations) in BVH
motion data. First, 3D positions (x-y-z coordinates) of all of the Automated Motion Data Processing for Dynamic
body joints in the BVH motion data are iteratively computed from Biomechanical Analysis in OpenSim
the root joint using local translations and rotations (i.e., a transfor- OpenSim (Delp et al. 2007) is a freely available software package
mation matrix) based on the predefined hierarchical structure of a that estimates biomechanical stresses including inertial forces
human skeleton in the BVH motion data. Then, the joint angles are exerted on human body joints due to changes in the velocity and
computed based on the vectors of bones between two connected direction of the motion (Anderson et al. 2012). Given the motion
body joints in a local coordinate system of the body, following and external force data, OpenSim performs inverse dynamics analy-
the definitions of horizontal, vertical, and rotational angles for sis with a multibody musculoskeletal system that has rigid skeletal
each body joint in 3D SSPP (Center for Ergonomics, University bones with virtual markers, as shown in Fig. 5, to calculate joint
of Michigan 2011). moments (Symeonidis et al. 2010). OpenSim is designed to conduct

Example of batch file Commands in batch file

General command lines • First line
3DSSPPBATCHFILE #
COM ----# - must be 3DSSPPBATCHFILE #
DES---- 0 “Example of Batch File“ #
AUT---- 1 # • COMMENT command
ANT---- 0 3 68.9 185 # - COM Anything you want to type to document your file. #
FRM-----1
FRM 1# Command lines for frame #1
• DESCRIPTION command
JOA----- 16.9698---- -7.5702----- 16.9698-- -7.5702---- 12.2084
----- -10.7369--- -20----- 0 -0.64244-- - -64.0564 - DES 0 (English unit) or 1 (Metric unit) "Task Name" #
----- -88.9659--- -85.4295 85.1361 -5.7582 2.2298-----
• AUTOEXPORT command
-4.3079 0----- 2.2298---- -4.3079---- 0.85105
----- -4.8446----- -20----- 0 ---- -23.9018 -80.3754
- AUT 0 (Do not export the results) or 1 (Export the results) #
----- -28.722-----
28 722 -81.1688
81 1688 93 9952
93.9952------7.998-----
7 998 88 3453
88.3453
• ANTHROPOMETRY Command
----- 0----- -4.0454----- 0----- 0----- ----- 0
----- 78.44----- -2.2991----- 1.3629---- 0----- ---- 0 #
- ANT 0 (Male) or 1 (Female) 3 (set according to the next two data)
HAN---- 20 -90 0 20 -90 0 # Height Weight #
EXP #
• FRAME Command
AUT 1#
- FRM Frame numbers. #
FRM 1 # Command lines for frame #2
JOA----- 18.0219- ----7.9502 ---- 18.0219---- -7.9502--- - 12.6864----- • SEGMENTANGLES Command
----- ----- -11.2829-- --20----- 0----- 0.90062-- --63.8582 - JOA 41 body segment angles (e.g., Hand left horizontal, hand left
----- -86.4393-- -84.6193 - - -87.5135---- -5.9056-- - 2.0483 vertical, hand left rotation, forearm left horizontal….) #
----- ----- -4.0814----- 0----- ----- 2.0483-- -4.0814 --- -0.30886
----- ----- -4.8004- -20----- -- - 0----- --------23.6184--- --80.5151 • HANDLOADS Command
28.331
----- ----- -28 331 -81
81.3217
3217- - -94
94.208
208---- -88.3682
3682- -- 87.8099
87 8099 - HAN Left magnitude angle left horizontal angle
magnitude, left vertical angle, angle, right
----- ----- 0----- --- -3.7762--- 0----- - 0----- ---- 0
magnitude, right vertical angle, right horizontal angle #
----- 78.6671---- -2.18---- 1.2585-- 0 0#
HAN---- 20 -90 0 20 -90 0 # • EXPORT Command
EXP--- - # - EXP (no data item, just for initiating data exports) #
AUT---- 1 #

(a) (b)

Fig. 4. Batch file to run 3D SSPP: (a) an example of a batch file; (b) commands in a batch file

© ASCE B4014005-5 J. Comput. Civ. Eng.

J. Comput. Civ. Eng., 2015, 29(4): B4014005

Downloaded from ascelibrary.org by Hong Kong Polytechnic University on 09/15/23. Copyright ASCE. For personal use only; all rights reserved.

Fig. 5. A screenshot of OpenSim window and a multibody model with virtual markers

biomechanical simulation with experimental data, such as marker segment—are determined using a subject’s height and weight based
positions and kinematics obtained from marker-based motion cap- on previous studies on these anthropometric parameters (Zatsiorsky
ture systems. For this reason, the TRC file format that contains et al. 1990; DeLeva 1996).
markers’ geometric information from optical motion capture sys- The next step is to generate a motion file (.mot) containing joint
tems such as VICON is the only motion data format available in angles from the BVH motion data. Because both the BVH motion
the current version of OpenSim. data and the motion file (.mot) in OpenSim define motions as joint
The procedures required to run OpenSim with marker data rotations in degrees relative to the initial position of the joint, geo-
(i.e., TRC file) are as follows (Anderson et al. 2012): (1) scaling metric information on skeleton structures from the BVH motion data
that adjusts both the mass properties (mass and moment-of-inertia) are immediately written to the motion file in OpenSim (.mot). The
and the dimensions of the body segment for the subject using multibody model that has motion information is shown in Fig. 7(b).
locations of markers, (2) inverse kinematics to create motions in
the body model by matching experimental markers with virtual
markers and to calculate joint angles, (3) inverse dynamics that Motion data External force
determines the net forces and torques at each joint that produces (.BVH) data
movement by solving the equations of motion with the given mo-
tion data (joint angles from inverse kinematics) and external force
data. For scaling (adjusting anthropometric factors) and inverse OpenSim Application Programming
kinematics (calculating body angles) processes, marker positions Interface (API)
in the TRC marker data are the primary sources; however, such
marker information is not available in the BVH motion data. To
Human body model Motion file
enable these two processes to be done with the BVH motion data, with anthropometric containing body
External load data
(.mot)
a user-friendly stand-alone system that automatically generates factors (.osim) angles (.mot)
joint moments from motion capture data was developed. This
system is based on the OpenSim API to generate a human multi-
Configuration file
body model (.osim) with anthropometric and physical properties (.xml)
(e.g., body mass, center of mass, and moment of inertia) fitted
to the subject, and a motion file (.mot) containing information
on joint angles at each body joint from the BVH motion data; Fig. 6 Inverse Dynamics Simulator
Provided by OpenSim
illustrates the overall workflow.
First, the proposed system creates a multibody model consisting
of body segments and joints based on the hierarchical structures of Joint moments (.sto)
bones and joints in the BVH motion data [Fig. 7(a)]. In addition,
anthropometric parameters of the multibody model—such as mass,
Fig. 6. Work flow for automated motion data processing in OpenSim
length, mass-center location, and moment-of-inertia of each body

© ASCE B4014005-6 J. Comput. Civ. Eng.

J. Comput. Civ. Eng., 2015, 29(4): B4014005

Downloaded from ascelibrary.org by Hong Kong Polytechnic University on 09/15/23. Copyright ASCE. For personal use only; all rights reserved.

Fig. 7. Multibody model from the BVH motion data: (a) a multibody model with anthropometric parameters fitted to the subject; (b) represented
motions in the multibody model based on the BVH motion data

Once the OpenSim body model (.osim) and the motion file lower leg were less than 5%. The radius of gyration is determined
(.mot) are generated, the system also creates a configuration file by the square root of the moment of inertia divided by the mass.
(.xml) that will be used by the OpenSim inverse dynamics simulator Considering that only the dynamic rotational moment is affected by
to integrate the model (.osim), motion (.mot), and external force the value of the moment of inertia, the error in the inertial param-
(.mot) files. The inverse dynamics simulator is an executable eters of a lower leg would not significantly affect the joint moments
module built from the source codes for inverse dynamics from at a knee joint. In addition, NRMSE values for body angles at
OpenSim, and thus enables us to perform inverse dynamics using elbows and knees were 0.079 and 0.081, respectively (Fig. 8).
the configuration file. The simulator saves the joint moments from These results indicate that the proposed approach accurately esti-
dynamic biomechanical analysis to a storage file (.sto). These mates anthropometric parameters and joint angles based on BVH
workflows are automatically processed only by inputting a sub- motion data, compared with the values from the existing approach.
ject’s anthropometric information (height and weight) and the
BVH motion data in the stand-alone system.
To verify the proposed approach, motion data was collected dur- An Experiment on Lifting Tasks
ing ladder climbing from one male subject using an optical motion
capture system, VICON, because climbing activities involve dy- The proposed automated motion data processing was experimen-
namic movements of the whole body. The raw data captured from tally tested by conducting an experiment on lifting tasks with pos-
VICON was converted into the motion data in different file formats: tural variations. An RGB-D sensor-based motion capture approach
TRC and BVH. Then, anthropometric parameters and joint angles was used to collect BVH motion data, and both static and dynamic
were compared from the approach that uses BVH motion data with biomechanical analyses in 3D SSPP and OpenSim were performed.
the parameters and angles from the existing approach of OpenSim The results of joint moment estimation by applying the proposed
that uses TRC motion data. To measure the differences in anthropo- approaches are presented and then compared with previous studies
metric parameters, the percentage error between the values was that estimated joint moments during lifting tasks using optical mo-
used. For joint angles, the normalized root-mean-square errors tion capture data in this section. In addition, the results from static
(NRMSE) were calculated between the values from this approach and dynamic biomechanical analyses are compared to understand
and the ones from the existing approach during one cycle of climb- how variations in postures and movements affect musculoskeletal
ing (240 frames, 2 s). As shown in Table 1, the differences in stresses. From the results, considerations when applying the pro-
anthropometric parameters except for the radius of gyration of a posed approaches for on-site biomechanical analysis and when

Table 1. Comparison of Anthropometric Parameters from OpenSim and the Proposed Approach
Anthropometric parameters Existing approach Our approach % of error
Upper arm Mass (kg) 1.96 1.97 −0.22
Length (m) 0.29 0.30 −5.43
Distance from center of mass to proximal joint as % of length 57.37% 57.72% −0.62
Radius of gyration as % of length, transverse 26.73% 26.89% 0.61
Radius of gyration as % of length, longitudinal 15.70% 15.80% 0.60
Radius of gyration as % of length, frontal 0.28 0.29 0.60
Lower leg Mass (kg) 3.50 3.30 5.71
Length (m) 0.43 0.42 2.30
Distance from center of mass to proximal joint as % of length 43.42% 44.58% −2.69
Radius of gyration as % of length, transverse 27.11% 25.10% −7.43
Radius of gyration as % of length, longitudinal 8.63% 10.20% 18.26
Radius of gyration as % of length, frontal 0.27 0.25 −9.89

© ASCE B4014005-7 J. Comput. Civ. Eng.

J. Comput. Civ. Eng., 2015, 29(4): B4014005

 Elbow Knee
from existing approach from our approach
120 120

100 100
Body Angle (degree)

80 80

60 60

40 40

20 20

0 0
0 50 100 150 200 0 50 100 150 200
Frames (120 frames per second)
Downloaded from ascelibrary.org by Hong Kong Polytechnic University on 09/15/23. Copyright ASCE. For personal use only; all rights reserved.

Frames (120 frames per second)

Fig. 8. Comparison of body angles between existing and proposed approaches

analyzing the results in the context of tasks are discussed in the next Moore 1992) [Fig. 9(a)]. After taking a break to minimize fatigue,
section. he was asked to apply the stoop technique in which the back is bent
to lift a block [Fig. 9(b)]. During the trials, Kinect took RGB-D im-
ages at a frame rate of 30 Hz, as shown at the top of Fig. 9, and the
Motion Data Collection images were processed in iPi Desktop Motion Capture to extract
Motion data during the concrete block lifting was collected by BVH motion data, as shown at the bottom of Fig. 9. The tracking
mimicking the tasks in a laboratory. A single RGB-D sensor was of motions was made through shape-fitting algorithms after match-
positioned at the front of a subject to minimize self-occlusion, and ing body silhouettes in T-pose images with 3D prior body models.
likewise was positioned 3 m away from the subject who wore slim However, hands are sometimes occluded by a concrete block during
clothes to minimize noise; Fig. 9 illustrates the experimental scenes. lifting, resulting in tracking errors. To clean up a sequence of incor-
A male subject (175 cm, 70 kg) was asked to stand in a T-pose, rect frames, inverse kinematics tools provided by iPi Desktop
facing the RGB-D sensor, and then repeatedly lift a 20-kg (196- Motion Capture were applied to estimate plausible arm postures
N) concrete block from one side on a floor and move it to the op- based on manipulated hand positions for occluded hands. In addi-
posite side 10 times. This protocol reflects practices during masonry tion, noise in motion data was reduced using post-processing filters.
work in which a worker lifts a block in stock, and puts it on a wall.
The RGB-D images associated with the T-pose were used just for
Results from Static and Dynamic Biomechanical
reference frames. To study an impact of postural variations on
Analyses
musculoskeletal stresses, two different lifting techniques were em-
ployed. First, he was asked to lift a block using the squat technique The BVH motion data was post-processed by applying the pro-
in which the back remains straight and the knees are bent (Garg and posed motion data conversion methods to obtain postural angles

Fig. 9. Motion data collection during concrete block lifting: (a) squat lifting; (b) stoop lifting

© ASCE B4014005-8 J. Comput. Civ. Eng.

J. Comput. Civ. Eng., 2015, 29(4): B4014005

 L5/S1 Disc Knee (left) Elbow (left)
Dynamic Static
300 120 40
JointMoments (Nm) 250 100 35
Squat Lifting
30
200 80 25
150 60 20
100 40 15
10
50 20 5
0 0 0
0.0 1.0 2.0 3.0 0.0 1.0 2.0 3.0 0.0 1.0 2.0 3.0
Time (s) Time (s) Time (s)
350 200 35
180
JointMoments (Nm)

300 30
160
Stoop Lifting
Downloaded from ascelibrary.org by Hong Kong Polytechnic University on 09/15/23. Copyright ASCE. For personal use only; all rights reserved.

250 140 25
200 120 20
100
150 80 15
100 60 10
40
50 20 5
0 0 0
0.0 1.0 2.0 3.0 0.0 1.0 2.0 3.0 0.0 1.0 2.0 3.0
Time (s) Time (s) Time (s)

Fig. 10. Biomechanical analysis results during squat and stoop lifting

for 3D SSPP and OpenSim. Anthropometric factors were adjusted However, in some cases, dynamic joint moments can be less than
based on the subject’s height and weight. To determine external static joint moments when deceleration exists. For example, at an
forces (e.g., hand and foot forces), it was assumed that the magni- elbow during squat lifting, dynamic joint moments were smaller
tude of external force exerted on each hand was 98 N, and that the than static joint moments in the middle of lifting (when the subject
direction of the forces was downward. In addition, the foot forces lifted a block at the highest height) because the lifting speed in a
were assumed as a sum of the weight of the subject and a concrete vertical direction decreased at this point, resulting in a deceleration
block. Based on these data, static and dynamic biomechanical that created joint moments in the opposite direction.
analyses were conducted to estimate joint moments during squat With regard to lifting techniques, the squat lifting is less stress-
and stoop lifting. ful to the back from both static and dynamic analyses in this case.
Fig. 10 shows joint moments at L5/S1 (i.e., an intervertebral As shown in Table 2, the peak and average joint moments at the L5/
disc between the fifth lumbar and first sacral vertebra), left knee, S1 disc during squat lifting are 91 and 84% of those during stoop
and left elbow joints during one cycle of squat and stoop lifting lifting in a static condition, and 84 and 66% in a dynamic condition.
(i.e., lift, carry, and put down a concrete block) from the static This result corresponds to the previous study that the squat lifting
and dynamic biomechanical analyses. In the graphs, the solid lines produced fewer maximum lumbar joint moments than the stoop lift-
indicate joint moments from dynamic biomechanical analysis while ing when a subject lifted the heavy object (15 kg; Hwang et al.
the dotted lines indicate joint moments from static biomechanical 2009). However, this result is not always the case because the stoop
analysis. Overall, the results show that joint moments from dy- lifting can maintain the object closer to the torso than squat lifting,
namic biomechanical analysis are higher than those from static resulting in the reduced moment arm of the load (Chaffin et al.
analysis. In a dynamic biomechanical model, the moment at a cer- 2006). For this reason, previous studies recommended squat lifting
tain body joint is defined as the sum of the static moment and the only when the subject can put an object between the feet to minimize
dynamic inertial forces (i.e., the instantaneous acceleration effect the moment arm of the load (van Dieën et al. 1999). In addition, the
due to the tangential rotation force and the rotational acceleration joint moments at a knee during stoop lifting are higher than those
effect; Chaffin et al. 2006). According to the study by McGill and during squat lifting, which is similar to the results from the previous
Norman (1985), the peak lumbar moment in a static condition was study (Hwang et al. 2009). No significant difference in elbow joint
84% of the peak lumbar moment in a dynamic condition during moments was observed in the results because the joint angles
lifting loads. The results are similar to the results from this study, between squat and stoop lifting were similar, as shown in Fig. 10.
by showing that the peak static joint moments at the L5/S1 disc are
79 and 73% of the peak dynamic joint moments at the L5/S1 disc
during squat and stoop lifting, respectively. One of the reasons why Discussion
the results showed higher dynamic joint moments compared with
The results from the experiment imply that the motion data-
static joint moments than the previous study is that the subject lifted
driven biomechanical analysis provides a robust measure of
a load from waist height in the previous study, whereas the subject
in the experiment lifted a concrete block from the floor. The lifting
speed when lifting a load (200 N) from the floor level could be Table 2. Comparison of Peak and Average Joint Moments during Squat
about 25% higher than the speed during lifting from waist height and Stoop Lifting
(Lavender et al. 2003). Thus, differences in lifting heights may con- L5/S1 disc Knee (left) Elbow (left)
tribute to differences in lifting speeds, resulting in higher dynamic
Joint moment (Nm) Peak Average Peak Average Peak Average
joint moments in the experiment than the ones from the previous
study. Overall results from the experiment indicate that rapid move- Squat lifting Dynamic 259.5 179.0 119.2 64.9 31.2 25.7
ments during lifting tasks could increase joint moments at the disc Static 205.3 163.0 80.8 42.5 33.7 27.7
by 25–30%. In addition, the acceleration effects also considerably Stoop lifting Dynamic 308.4 271.8 207.6 166.1 32.9 23.4
Static 225.6 193.2 87.9 49.5 24.8 17.1
increase joint moments at other joints, such as a knee and an elbow.

© ASCE B4014005-9 J. Comput. Civ. Eng.

J. Comput. Civ. Eng., 2015, 29(4): B4014005

Downloaded from ascelibrary.org by Hong Kong Polytechnic University on 09/15/23. Copyright ASCE. For personal use only; all rights reserved.

Fig. 11. Comparison of a trunk flexion angle in (a) 3D SSPP; (b) BVH motion data

musculoskeletal stress from both static and dynamic points of view, errors in the trunk flexion angle that is calculated from the
given motion capture data (e.g., BVH files). From these results, it BVH motion data [the bottom of Fig. 11(b)], resulting in slightly
was also found that musculoskeletal stresses vary depending on the different postures in 3D SSPP. Generally, the hierarchical structures
acceleration of motions and postures during certain tasks. In con- of bones and joints in the BVH motion data vary depending on the
struction, working speeds and postures are affected by diverse fac- type of motion capture system and algorithm. For this reason, when
tors, for example, individual factors (e.g., personal preference), applying the BVH motion data to 3D SSPP, one should consider
environmental factors (e.g., working spaces), or managerial factors the differences in skeleton structures between skeleton models in
(e.g., work schedule). Thus, it is concluded that on-site biome- 3D SSPP and motion data, and adjust them if the differences are
chanical analysis is required to consider these factors that may significant.
expose workers to different levels of WMSD risks even when per- The reliability and practicability of the proposed on-site biome-
forming the same types of tasks in construction. In addition, the chanical analysis during construction tasks relies on not only auto-
usability issue of the proposed method is also important from mated motion data processing, but also on motion data collection.
the practical perspective in making the biomechanical models use- Previous studies found that vision-based motion data has some er-
ful when performing ergonomic evaluations during tasks (Chaffin rors in body angles (about 10°; Han et al. 2012a) and bone
1997). In this context, it should be clear that the automatic proc- lengths (about 6.3 cm; Han et al. 2013b). Especially, occlusions
esses of motion data obtained directly from work places enable (e.g., self-occlusion or occlusion by an object) significantly in-
ergonomists and practitioners to identify a potential risk of WMSDs creased errors in motion data, resulting in a failure to track the
that exists in certain tasks and work environments by evaluating human body (Han et al. 2012a). To address the occlusion issue,
hazardous internal loading conditions in a timely manner without Han et al. (2012a) suggested the need for positioning sensing de-
technical sophistication or skill. vices at optimal viewpoints and using multiple sensors to minimize
Notably, in-depth understanding of the differences in body con- the impact of occlusions. In addition, when certain body joints
figurations between the BVH motion data and the 3D SSPP body are lost or incorrectly detected in a relatively short period of time
model may lead to the further improvement on the accuracy of bio- due to occlusions, errors in vision-based motion data can be re-
mechanical analysis. Skeleton structures of the multibody model in duced through post processing on motion data (such as trajectory
OpenSim follow those in the BVH motion data excluding joints in recovery) that optical motion capture systems apply to estimate
hands, and thus the motions can be exactly simulated according to missing markers (Sturman 1994). However, vision-based motion
the motions in the BVH motion data. On the other hand, 3D SSPP capture approaches rely only on pixel information on 2D or 3D
has its own skeleton structures that are different from those in the images that make it fundamentally difficult to provide motion data
BVH motion data, which makes errors when calculating joint an- that is as accurate as the data from optical motion capture systems.
gles. For example, in 3D SSPP, the trunk flexion angle is defined as To provide the reliability of on-site biomechanical analysis, the
an angle between the projection of the trunk-axis (the center of the acceptable errors in motion data for biomechanical analysis need
hips to the center of the shoulders) and the positive Y-axis as shown to be further studied through sensitivity analysis by varying body
at the bottom of Fig. 11(a). In the top view of the skeleton model in angles. In addition, the use of infrared RGB-D sensors such as a
3D SSPP [the top of Fig. 11(a)], it is found that hips and shoulders Kinect for collecting 3D images is limited to only indoor environ-
are aligned. On the other hand, the hips are located slightly forward ments with relatively small workspaces in construction sites. A
in the Y-axis, compared with the shoulders [the top of Fig. 11(b)]. multiple camera-based motion capture approach is expected (Han
The differences in skeleton structures and joint positions cause et al. 2012b, 2013a) to overcome the limitation of RGB-D sensors.

© ASCE B4014005-10 J. Comput. Civ. Eng.

J. Comput. Civ. Eng., 2015, 29(4): B4014005

 However, the feasibility of on-site biomechanical analysis using data. In addition, for OpenSim, a multibody model was created
vision-based motion data needs to be further studied by conducting with anthropometric parameters adjusted for a subject based on
case studies in the diverse conditions of construction sites. the hierarchical structures of bones and joints in the BVH motion
When applying biomechanical analysis on construction tasks, data, and computed joint angles based on joint rotations in degrees
however, the selection of adequate analysis should be made con- that are available in the BVH motion data. The proposed motion
sidering the purpose of the analysis due to differences in tolerance data processing for OpenSim was verified by comparing anthropo-
limits under static and dynamic conditions. The results from the metric parameters and joint angles from this approach with those
experiments indicated that acceleration effect could increase joint from the existing approach in OpenSim.
moments up to 30% in lifting tasks. Therefore, static analysis using In addition, an experiment on lifting tasks was conducted
3D SSPP is appropriate for tasks involving slow motions where not only to test the feasibility of the proposed motion data process-
accelerations can be ignored, whereas tasks involving jerking ing, but also to understand the differences between static and
motions require dynamic biomechanical analysis in OpenSim. dynamic biomechanical analyses on construction tasks with diverse
Downloaded from ascelibrary.org by Hong Kong Polytechnic University on 09/15/23. Copyright ASCE. For personal use only; all rights reserved.

However, it should be noted that there is no available threshold postural variations. The results from the experiment showed that
to determine whether the dynamic joint moments are hazardous the proposed approaches for motion data processing were
or not. Tissue injuries occur when the applied musculoskeletal successfully used to perform static and dynamic biomechanical
stresses exceed the failure tolerance referring to the strength of analyses by showing similar results from previous studies. From
the tissue (McGill 1997). Because it is difficult to specify individual the experiment, it was also found that the speed of motions
differences in joint strength, population-based data is generally and postures could significantly affect the magnitude of musculo-
used to determine hazardous internal loads. For example, 3D SSPP skeletal stresses even though workers perform the same tasks,
compares the joint moments produced at various body joints during which implies that on-site biomechanical analysis is necessary
tasks with the static strength moments reported from studies of to reflect postural variations during the performance of construction
various populations performing different types of exertions by set- tasks.
ting the maximum permissible limit as the joint strength that To enhance the reliability and practicability of on-site biome-
only 25% of men and 1% of women can exert (Center for Ergo- chanical analysis using motion data from vision-based approaches,
nomics, University of Michigan 2011). However, because dynamic understanding body configurations in the BVH motion data is re-
strengths are more complex than static strengths, studies on dy- quired to reduce errors in biomechanical analysis, and the accuracy
namic joint strengths have not yet been fully conducted (Chaffin of motion data from vision-based motion capture approaches needs
et al. 2006). This means that joint moments from dynamic biome- to be improved. In addition, when practitioners apply on-site
chanical analysis cannot be evaluated to determine the degree of biomechanical analysis during construction tasks, the selection of
risk in a given population of workers, but can only be relatively static or dynamic biomechanical analysis should be carefully made
compared. depending on the types of the tasks due to the effect of the accel-
When assessing WMSD risks through biomechanical analysis, eration of motions. The identification of excessive musculoskeletal
ergonomic evaluations and interventions should not be made frag- stresses through biomechanical analysis will help to determine an
mentarily only focusing on postural risks that create excessive in- ergonomic intervention for eliminating the risk of WMSDs. How-
ternal forces on certain body parts. For example, the stoop lifting ever, practitioners need to consider not only postural risks identified
technique is generally recommended because frequent back bend- from biomechanical analysis, but also the design of tasks when
ing can cause back pain and even back injuries. However, because developing effective interventions.
squat lifting requires higher joint moments in lower extremities and It is expected that the proposed approach will help practitioners
higher energetic demand (e.g., oxygen consumption; Duplessis perform on-site biomechanical analysis in a practical manner
et al. 1998), it was reported that subjects tended to change their without technical sophistication or skill, given motion data from
lifting technique from squat to stoop lifting during repetitive vision-based approaches. On-site biomechanical analysis has great
lifting to avoid or diminish fatigue development (Resnick 1996; potential as a field-based ergonomic risk assessment method
Fogleman and Smith 1995). In this case, ergonomic interventions that determines the degree of risk on each body part that other
should be made considering not only postural risks due to individ- ergonomic evaluation methods (e.g., self-reports, observational
ual preference, but also other factors such as the horizontal methods, and direct measurements) cannot provide. Thus, it can
and vertical position of the load and the load mass (van Dieën identify potentially hazardous construction tasks, contributing to
et al. 1999). minimizing the risks of WMSDs by providing behavioral feedback
to workers or redesigning work processes and environments that
may cause excessive musculoskeletal stresses. Ultimately, the con-
Conclusions tinuous monitoring of biomechanical stresses during construction
tasks will enhance the understanding of the gap between physical
Motion data-driven biomechanical analysis was proposed during work demands and workers’ capability, and offer a firm foundation
construction tasks using motion data obtained from vision-based for the improvement of workers’ productivity, as well as health in
motion capture approaches. Specifically, the differences of configu- construction.
rations in BVH motion data from vision-based motion capture and
data types required for existing biomechanical analysis tools, 3D
SSPP and OpenSim, was studied and compared. It was found that: Acknowledgments
(1) definitions of body angles in BVH motion data are different
from the definitions of body angles in 3D SSPP, and (2) experimen- The work presented in this paper was supported financially with a
tal marker positions that are not available in BVH motion data are National Science Foundation Award (No. CMMI-1161123) and a
essential for generating anthropometric parameters and joint angles CPWR grant through NIOSH cooperative agreement OH009762.
required to compute musculoskeletal stresses in OpenSim. To ad- Any opinions, findings, and conclusions or recommendations ex-
dress these issues, body angles required for 3D SSPP are computed pressed in this paper are those of the authors and do not necessarily
directly from positions of body joints generated from BVH motion reflect the views of the National Science Foundation.

© ASCE B4014005-11 J. Comput. Civ. Eng.

J. Comput. Civ. Eng., 2015, 29(4): B4014005

 References Gatti, U. C., Migliaccio, G. C., Schneider, S., and Fierro, R. (2010).
“Assessing physical strain in construction workforce: A first step for
Abásolo, L., et al. (2012). “Prognostic factors for long-term work improving safety and productivity management.” Proc., 27th Int. Symp.
disability due to musculoskeletal disorders.” Rheumatology Int., 32(12), on Automation and Robotics in Construction (ISARC), International
3831–3839. Association for Automation and Robotics in Construction (IAARC),
Albers, J. T., and Estill, C. F. (2007). Simple solutions: Ergonomics for 255–264.
construction workers, National Institute for Occupational Safety and Han, S., Achar, M., Lee, S., and Peña-Mora, F. (2012a). “Automated 3D
Health (NIOSH), Cincinnati, OH. human skeleton extraction using range cameras for safety action sam-
Alwasel, A., Elrayes, K., Abdel-Rahman, E., and Haas, C. (2012). “Reduc- pling.” Proc., 12th Int. Conf. on Construction Applications of Virtual
ing shoulder injuries among construction workers.” Proc., Joint Conf. of Reality (ConVR), National Taiwan University Press, Taiwan.
the 8th World Conf. of the Int. Society for Gerontechnology (ISG) and Han, S., Lee, S., and Armstrong, T. (2013a). “Automated 3D skeleton
the 29th Int. Symp. on Automation and Robotics in Construction extraction from multiple videos for biomechanical analysis on sites.”
(ISARC), International Association for Automation and Robotics in Proc., 8th Int. Conf. on Prevention of Work-Related Musculoskeletal
Downloaded from ascelibrary.org by Hong Kong Polytechnic University on 09/15/23. Copyright ASCE. For personal use only; all rights reserved.

Construction (IAARC), 270–274. Disorders (PREMUS), Pusan National Univ. School of Medicine,
Aminian, K., and Najafi, B. (2004). “Capturing human motion using South Korea.
body-fixed sensors: Outdoor measurement and clinical applications.” Han, S., Lee, S., and Peña-Mora, F. (2012b). “Vision-based motion detec-
Comput. Anim. Virtual Worlds, 15(2), 79–94. tion for safety behavior analysis in construction.” Proc., 2012 Construc-
Anderson, F. C., et al. (2012). “OpenSim user’s guide.” 〈https://opensim tion Research Congress (CRC), ASCE, Reston, VA, 21–23.
.stanford.edu〉 (April 26, 2014). Han, S., Lee, S., and Peña-Mora, F. (2013b). “Vision-based detection of
Armstrong, T., et al. (1996). “Musculoskeletal disorders: Work-related risk unsafe actions of a construction worker: Case study of ladder climbing.”
factors and prevention.” Int. J. Occup. Environ. Health, 2(3), 239–246. J. Comput. Civil. Eng., 10.1061/(ASCE)CP.1943-5487.0000279,
Boschman, J., van der Molen, H., Sluiter, J., and Frings-Dresen, M. (2012). 635–644.
“Musculoskeletal disorders among construction workers: A one-year Hwang, S., Kim, Y., and Kim, Y. (2009). “Lower extremity joint kinetics
follow-up study.” BMC Musculoskel. Disord., 13(1), 196. and lumbar curvature during squat and stoop lifting.” BMC Musculos-
Bureau of Labor Statistics (BLS). (2012). “National census of fatal occupa- kel. Disord., 10, 15.
tional injuries in 2011.” 〈http://www.bls.gov/news.release/cfoi.nr0.htm〉 Inyang, N., Al-Hussein, M., El-Rich, M., and Al-Jibouri, S. (2012). “Ergo-
(Apr. 26, 2014). nomic analysis and the need for its integration for planning and assess-
Center for Construction Research, and Training (CPWR). (2013). “The ing construction tasks.” J. Constr. Eng. Manage., 10.1061/(ASCE)CO
construction chart book: The U.S. construction industry and its workers .1943-7862.0000556, 1370–1376.
(fifth edition).” 〈http://www.cpwr.com/publications/construction-chart Jäger, M., and Luttmann, A. (1999). “Critical survey on the biomechanical
-book〉 (Apr. 26, 2014). criterion in the NIOSH method for the design and evaluation of manual
Center for Ergonomics, University of Michigan. (2011). 3D static strength lifting tasks.” Int. J. Ind. Ergon., 23(4), 331–337.
prediction program: User’s manual, Univ. of Michigan, Ann Arbor, MI.
Jia, B., Kim, S., and Nussbaum, M. A. (2011). “An EMG-based model to
Chaffin, D. B. (1997). “Development of computerized human static
estimate lumbar muscle forces and spinal loads during complex, high-
strength simulation model for job design.” Hum. Factors Ergon.
effort tasks: Development and application to residential construction
Manuf., 7(4), 305–322.
using prefabricated walls.” Int. J. Ind. Ergon., 41(5), 437–446.
Chaffin, D. B., Andersson, G. B., and Martin, B. J. (2006). Occupational
Kim, S., Nussbaum, M. A., and Jia, B. (2011). “Low back injury risks
biomechanics, 4th Ed., Wiley, Hoboken, NJ.
during construction with prefabricated (panelised) walls: Effects of task
Chun, C. W., Jenkins, O. C., and Mataric, M. J. (2003). “Markerless kin-
and design factors.” Ergonomics, 54(1), 60–71.
ematic model and motion capture from volume sequences.” Proc., 2003
Kumar, S. (1999). Biomechanics in ergonomics, Taylor & Francis,
IEEE Computer Society Conf. on Computer Vision and Pattern Recog-
Philadelphia.
nition (CVPR), IEEE, Los Alamitos, CA.
Lavender, S. A., Andersson, G. B., Schipplein, O. D., and Fuentes, H. J.
Corazza, S., Mündermann, L., Chaudhari, A. M., Demattio, T., Cobelli, C.,
and Andriacchi, T. P. (2006). “A markerless motion capture system (2003). “The effects of initial lifting height, load magnitude, and lifting
to study musculoskeletal biomechanics: Visual hull and simulated speed on the peak dynamic L5/S1 moments.” Int. J. Ind. Ergon., 31(1),
annealing approach.” Ann. Biomed. Eng., 34(6), 1019–1029. 51–59.
Dai, F., and Ning, X. (2013). “Remote sensing enabling technologies for Lee, M. W., and Cohen, I. (2006). “A model-based approach for estimating
identification and evaluation of construction worker’s musculoskeletal human 3D poses in static images.” Pattern Anal. Mach. Intell., 28(6),
disorder risks: A review and future.” Proc., 30th Int. Symp. on Automa- 905–916.
tion and Robotics in Construction and Mining (ISARC), International Li, G., and Buckle, P. (1999). “Current techniques for assessing physical
Association for Automation and Robotics in Construction (IAARC). exposure to work-related musculoskeletal risks with emphasis on
David, G. C. (2005). “Ergonomic methods for assessing exposure to risk posture-based methods.” Ergonomics, 42(5), 674–695.
factors for work-related musculoskeletal disorders.” Occup. Med., Marras, W. S., and Radwin, R. G. (2005). “Biomechanical modeling.” Rev.
55(3), 190–199. Hum. Factors Ergon., 1(1), 1–88.
Davis, R. B., Ounpuu, S., Tyburski, D., and Gage, J. R. (1991). “A gait McGill, S. M. (1997). “The biomechanics of low back injury: Implications
analysis data collection and reduction technique.” Hum. Mov. Sci., on current practice in industry and the clinic.” J. Biomech., 30(5),
10(5), 575–587. 465–475.
DeLeva, P. (1996). “Adjustments to Zatsiorsky-Seluyanov’s segment inertia McGill, S. M., and Norman, R. W. (1985). “Dynamically and statically
parameters.” J. Biomech., 29(9), 1223–1230. determined low back moments during lifting.” J. Biomech., 18(12),
Delp, S. L., et al. (2007). “OpenSim: Open-source software to create and 877–885.
analyze dynamic simulations of movement.” Biomed. Eng., 54(11), Meerding, W. J., Ijzelenberg, W., Koopmanschap, M. A., Severens, J. L.,
1940–1950. and Burdorf, A. (2005). “Health problems lead to considerable produc-
Duplessis, D. H., et al. (1998). “Effect of semi-rigid lumbosacral orthosis tivity loss at work among workers with high physical load jobs.” J. Clin.
use on oxygen consumption during repetitive stoop and squat lifting.” Epidemiol., 58(5), 517–523.
Ergonomics, 41(6), 790–797. Meredith, M., and Maddock, S. (2001). Motion capture file formats
Fogleman, M., and Smith, J. L. (1995). “The use of biomechanical mea- explained, Dept. of Computer Science, Univ. of Sheffield, U.K.
sures in the investigation of changes in lifting strategies over extended Mitropoulos, P., and Memarian, B. (2013). “Task demands in masonry
periods.” Int. J. Ind. Ergon., 16(1), 57–71. work: Sources, performance implications, and management strategies.”
Garg, A. R. U. N., and Moore, J. S. (1992). “Prevention strategies and the J. Constr. Eng. Manage., 10.1061/(ASCE)CO.1943-7862.0000586,
low back in industry.” Occup. Med., 7(4), 629–640. 581–590.

© ASCE B4014005-12 J. Comput. Civ. Eng.

J. Comput. Civ. Eng., 2015, 29(4): B4014005

 Moeslund, T. B., and Granum, E. (2001). “A survey of computer vision- Schneider, S. P. (2001). “Musculoskeletal injuries in construction: A review
based human motion capture.” Comput. Vis. Image Understand., 81(3), of the literature.” Appl. Occup. Environ. Hyg., 16(11), 1056–1064.
231–268. Shotton, J., et al. (2013). “Real-time human pose recognition in parts from
Moeslund, T. B., Hilton, A., and Krüger, V. (2006). “A survey of advances single depth images.” Commun. ACM, 56(1), 116–124.
in vision-based human motion capture and analysis.” Comput. Vis. Siddiqui, M., and Medioni, G. (2010). “Human pose estimation from a sin-
Image Understand., 104(2), 90–126. gle view point, real-time range sensor.” Proc., 2010 IEEE Computer
Monroe Keyserling, W., and Budnick, P. M. (1987). “Non-invasive meas- Society Conf. on Computer Vision and Pattern Recognition Workshops
urement of three dimensional joint angles: Development and evaluation (CVPRW), IEEE Computer Society, Los Alamitos, CA, 1–8.
of a computer-aided system for measuring working postures.” Int. J. Simoneau, S., St-Vincent, M., and Chicoine, D. (1996). “Work-related
Ind. Ergon., 1(4), 251–263. musculoskeletal disorders (WMSDs)—A better understanding for
Morris, J. R. W. (1973). “Accelerometry—A technique for the measure- more effective prevention.” Technical Guide RG-126-ang, IRSST,
ment of human body movements.” J. Biomech., 6(6), 729–736. Montréal, Canada.
National Institute for Occupational Safety, and Health (NIOSH). (1981). Sturman, D. J. (1994). “A brief history of motion capture for computer char-
Downloaded from ascelibrary.org by Hong Kong Polytechnic University on 09/15/23. Copyright ASCE. For personal use only; all rights reserved.

acter animation.” Course Notes 9, Character Motion Systems, ACM

“Work practices guide for manual lifting.” Technical Rep. No:
SIGGRAPH 1994, Los Angeles.
81–122, U.S. Dept. of Health and Human Services, Cincinnati, OH.
Symeonidis, I., Kavadarli, G., Schuller, E., and Peldschus, S. (2010).
Paul, J. A., and Douwes, M. (1993). “Two-dimensional photographic
“Simulation of biomechanical experiments in OpenSim.” Proc., 12th
posture recording and description: A validity study.” Appl. Ergon.,
Mediterranean Conf. on Medical and Biological Engineering and
24(2), 83–90.
Computing (MEDICON), International Federation for Medical and Bio-
Plagemann, C., Ganapathi, V., Koller, D., and Thrun, S. (2010). “Real-time logical Engineering (IFMBE), Germany, 107–110.
identification and localization of body parts from depth images.” Proc., van Dieën, J. H., Hoozemans, M. J., and Toussaint, H. M. (1999). “Stoop or
2010 IEEE Int. Conf. on Robotics and Automation (ICRA), IEEE, squat: A review of biomechanical studies on lifting technique.” Clin.
Piscataway, NJ, 3108–3113. Biomech., 14(10), 685–696.
Poppe, R. (2007). “Vision-based human motion analysis: An overview.” Wiles, A. D., Thompson, D. G., and Frantz, D. D. (2004). “Accuracy
Comput. Vis. Image Understand., 108(1), 4–18. assessment and interpretation for optical tracking systems.” Proc., SPIE
Radwin, R. G., Marras, W. S., and Lavender, S. A. (2001). “Biomechanical Medical Imaging 2004: Visualization, Image-Guided Procedures,
aspects of work-related musculoskeletal disorders.” Theor. Issues and Display, International Society for Optics and Photonics (SPIE),
Ergon. Sci., 2(2), 153–217. Bellingham, Washington, DC, 421–432.
Ray, S. J., and Teizer, J. (2012). “Real-time construction worker posture Zatsiorsky, V. M., Seluyanov, V. N., and Chugunova, L. G. (1990).
analysis for ergonomics training.” Adv. Eng. Inform., 26(2), 439–455. “Methods of determining mass-inertial characteristics of human body
Resnick, M. (1996). “Postural changes due to fatigue.” Comput. Ind. Eng., segments.” Contemporary problems of biomechanics, G. G. Chemyi,
31(1), 491–494. and S. A. Regirer, ed., CRC Press, MA, 272–291.

© ASCE B4014005-13 J. Comput. Civ. Eng.

J. Comput. Civ. Eng., 2015, 29(4): B4014005