Computer-Aided Design and Applications

ISSN: (Print) 1686-4360 (Online) Journal homepage: http://www.tandfonline.com/loi/tcad20

A method for posture prediction of the upper

trunk of video terminal operators

Luca Di Angelo & Paolo Di Stefano

To cite this article: Luca Di Angelo & Paolo Di Stefano (2017) A method for posture prediction
of the upper trunk of video terminal operators, Computer-Aided Design and Applications, 14:1,
28-37, DOI: 10.1080/16864360.2016.1199753

To link to this article: http://dx.doi.org/10.1080/16864360.2016.1199753

Published online: 20 Jun 2016.

Submit your article to this journal

Article views: 11

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at

http://www.tandfonline.com/action/journalInformation?journalCode=tcad20

Download by: [Ryerson University Library] Date: 06 October 2016, At: 06:47
COMPUTER-AIDED DESIGN & APPLICATIONS, 2017
VOL. 14, NO. 1, 28–37
http://dx.doi.org/10.1080/16864360.2016.1199753

A method for posture prediction of the upper trunk of video terminal operators
Luca Di Angelo and Paolo Di Stefano
University of L’Aquila, Italy

ABSTRACT KEYWORDS
This paper presents a method for posture prediction of the upper trunk of video terminal (VDT) oper- Posture prediction;
ators, which is then verified by means of some test cases. The prediction of the upper trunk posture ergonomics; multi-body
is, in fact, a very difficult task to carry out due mainly to the complexity of the anatomy of the spine virtual human models;
and the surrounding muscles. The method being proposed in this paper is based on the integration artificial neural network
of the knowledge which is obtained experimentally through the posture analysis of real cases into a
configured human multi-body kinematic model which has been implemented in a commercial CAD
system. A trained artificial neural network retains the knowledge concerning the VDT operator’s pos-
tures detected in different working positions. The posture simulations obtained with the proposed
method are subsequently compared with the real ones determined by a 3D scanner. The results
obtained confirm the effectiveness of such a method, which is deemed promising to implement
other anthropometric data and further human poses.

1. Introduction The spine, on its part, can be represented as a complex

Every individual assumes a posture (for the most part, multi-joint structure whose configuration determines the
without a conscious decision) when interacting with upper-trunk posture. However, the considerable com-
the surrounding workplace. Such adaptability is possible plexity of the articular system of the spine and its inter-
thanks to the complex anatomy of the musculoskele- action with the head make it difficult to model the upper
tal system, which allows of different degrees of free- trunk without large approximations ([14] and [19]). Con-
dom. However, not all of postures are necessarily correct. sidering that there is not a unique relation between pos-
When a non-correct posture is adopted for a long period ture and human interaction with environment, but rather
of time it may cause musculoskeletal injury; as an exam- a complex one, which is determined by unpredictable
ple, low levels of seating comfort often lead to muscu- kinematic constraints (such as the fibrous connective tis-
loskeletal complaints such as low back pain (LBP) [15]. In sues that join the articular surfaces of bones), it follows
order to correctly design the working place, human fac- that human pose, and hence spine configuration, are dif-
tor analysis is an activity that should not be disregarded. ficult to reproduce, due precisely to the postural attitudes
Human models for ergonomic design are implemented of each individual and the voluntary human intents.
in many commercial CAD systems [11]. For one thing, In order to solve the inverse kinematic model, each
using a Digital Human Model (DHM) in designing a part of the human body can be described by a prop-
new working place arrangement or industrial product erly defined biomechanical equation which characterizes
may reduce both the costs and the time required to pro- some aspects of human performance, such as discomfort
duce and test product prototypes ([3], [12] and [16]). or required energy [17]. In this sense, human posture is
Another reason for favoring the use of DHM has to do deduced by solving an equation system ([14], [9], [10], [8]
with the highly variable level of human performance and and [18]). In the related literature, this approach is used
attributes, which necessarily require a greater number of in simple cases, for which few biomechanical equations
tests for a complete analysis. The reliability of a human are required, such as, for instance, the case of the right
posture simulation, therefore, lies on the capability of the upper limb or arm. This affords a very limited number of
DHM to reproduce the human body performance, as well degrees of freedom, though, especially if we bear in mind
as the complex interactions between the subject and the the fact that the articular system of the spine requires at
external environment. least 24 DOF to be described.

CONTACT Luca Di Angelo [email protected]

© 2016 CAD Solutions, LLC, http://www.cadanda.com

 COMPUTER-AIDED DESIGN & APPLICATIONS 29

Another way of implementing a complex inverse kine- and the thoracic vertebrae. The lumbar vertebrae form a
matic model is by using artificial neural networks (ANN). curvature, called lordosis, which is posteriorly concave.
Artificial neural networks, like biological neural net- The thoracic vertebrae, on their part, form a convex cur-
works, contain a collection of neuron units communicat- vature, called kyphosis (figure 1). These curvatures are
ing with each other via axon connections. The memory of typical of the standing posture of any human being. In
a neural network is included in the synaptic weights that the sitting posture, on the contrary, the lordosis curva-
are adaptively trained by a learning mechanism. A multi- ture generally tends to be less curved and, in some cases,
layer feed-forward neural network with an arbitrarily may even disappear [5].
large number of units in hidden layers can approximate
any real continuous function. It is for this reason that
neural networks are also proposed to implement human
biomechanical models ([11], [13] and [20]). Rezzoug and
Gorce [13], on the one hand, present a technique based
on neural networks that maps the fingertip 3D position
and the corresponding joint angles so as to predict the
hand and finger postures during grasping tasks. Zhang
et al. [20], on the other hand, present the concept and the
implementation of the ANN-based posture transforma-
tion methodology, which reconstructs the configuration
of the human body, in arbitrary postures, through 27
landmarks. More recently, Bataineh et al. [2] have pro-
posed the use of an ANN for predicting the upper-body
posture for a 41-DOF human model. This is a prelim-
inary work to test the capability of a neural network
to reproduce the inverse kinematics model of a digital
manikin. They do not introduce nothing news concern-
ing the reproduction of a realistic posture.
In order to perform an accurate prediction the upper-
trunk posture of real video terminal operators, in this
paper a new approach is proposed, which integrates the
kinematic model of the spine with the knowledge about
Figure 1. Human spine.
human postures. With the intent to survey the knowledge
concerning human postures, who are somewhat con-
strained by the assigned working positions, an accurate The trunk posture, therefore, can be defined in terms
method to detect the shape of the spine has been used. of the orientation of the motion segments and in terms
Then the data related to the environment interaction of the orientation of the thorax with respect to the
has been transformed into a map between upper trunk pelvis. The motion segments are the fundamental build-
kinematic parameters and manikin interaction with the ing blocks of the spine and consist of two vertebrae
environment. This map defines the configuration of the connected by intervertebral discs and by a number of
multi-body structure representing a virtual manikin. ligaments. 24 motion segments which correspond to ver-
tebrae can be identified in the human spine.
2. Trunk anatomy
3. Enhanced digital human model
The geometry of the lumbar spinal column plays an
important role in determining human posture. It is a The method proposed in this paper is briefly described
complex anatomical structure that interacts with the sur- in the chart in figure 2. It performs an Enhanced Digi-
rounding musculature and that is constrained by the tal Human Model (EDHM), based on a classic kinematics
position of arms, head and pelvis. The lumbar spine is a model of the human body, in which the knowledge con-
flexible structure composed of vertebrae, which are rigid cerning the postural attitude is implemented by means
elements. It consists of 24 vertebrae (figure 1): 5 in the of an Artificial Neural Network (ANN). The number of
lumbar part (l1÷l5), 12 in the thoracic part (t1÷t15), the degree of freedom of a manikin is greater than the
and 7 in the cervical part (c1÷c7). The trunk posture constraints that a working plane can determine. So that,
is mainly conditioned by the orientation of the lumbar for a given manikin position, the redundancy of degrees
30 L. DI ANGELO AND P. DI STEFANO

Figure 2. The proposed method chart.

of freedom (DOF) involves that alternative postures can which works performing a specific activity in a specific
be assumed. For each unconstrained articulation, it is working place.
known that there is a defined range of possible move-
ments [1]; within this range, the specific value of these
parameters determines a specific posture. The posture
3.1. Posture attitude knowledge acquisition
assumed by a specific subject can be described by a proper
function which relates articular configuration with the In order to ascertain the postural attitude of VDT oper-
tasks in the workplace, it is possible to say that this func- ators, some experiments have been conducted, and a
tion describes the specific postural attitude of the subject. pattern has been defined for setting up different work-
Typically in commercial DHM, the redundancy in kine- ing environment configurations. The number of postures
matic parametric specification are solved according to a which are necessary to identify the data to train the pos-
criterion which is generally unknown to the user. This ture simulator grows factorially with the number of DOF.
criterion makes manikin postures rigid, un-personal and In this study, the posture simulator has been trained to
unnatural. reproduce postures that are defined in the sagittal plane;
In the Enhanced Digital Human Model, here pro- such is the case for the VDT operator. For this purpose, a
posed, the under-determinate parameters of the kine- configurable working environment has been reproduced
matic model are furnished by a properly trained Artificial in the laboratory (figure 3).
Neural Network that controls each kinematics segment The working environment chosen consists of a seat
when the manikin interacts with the working place. The with a footrest, a table and a target to be observed by the
ANN is, actually, the component of the model which subject selected for the experiments. All the components
retains the knowledge concerning the relations between of the simulated working environment may, nonetheless,
environment interaction and body part configuration. In be modified to define different working configurations.
order to furnish the specific posture attitude of a real sub- The environmental parameters used to design the exper-
ject, some experiments have been performed in which iment comply with EN ISO 9241 (2001); they are the
the posture of VDT operators in action are analyzed. The following (figure 4):
EDHM consists in two main parts: the Digital Human
Model (DHM) and the Artificial Neural Network which • θ = angle between the line of view and the horizontal
retains the specific postural attitude. When the anthropo- plane;
metric data are assigned, the DHM is configured to repro- • a = distance between the buttocks and the table edge;
duce the dimension of a specific subject. The ANN is • b = height of the table with respect to the seat;
trained with the data concerning specific VDT operators • c = height of the seat with respect to the footrest.
 COMPUTER-AIDED DESIGN & APPLICATIONS 31

acquisition of the back needs to be complete and per-

formed across the vertebral column. After the acquisition
is completed, the point cloud is processed by a typical
smoothing operation and then tessellated. This filtering
process is useful to reduce outliers and large noise.
First of all, however, before scanning the geometry of
the back, markers are placed on the individual in corre-
spondence with the landmarks T1 and L5 of the spine
column (figure 5a) so as to identify its beginning (ver-
tebra T1) and its end (vertebra L5). Second, and with a
view to detecting the vertebral column, a specific method
to evaluate the symmetry line is used [6], which performs
a 3D virtual reconstruction of the spine. Then, the typical
human global reference frame is estimated, whose char-
Figure 3. The experimental set-up. acteristic planes are coronal (C ), transversal (T ) and
sagittal (S ).
For the type of postures which have been analysed, the
symmetry line is nominally contained in the sagittal plane.
In real cases, however, the symmetry line detected for a
VDT operator is a curve which deviates from the sagittal
plane. This deviation is, at any rate, limited, and, conse-
quently, there is not a great difference between the 3D
θ
symmetry line and its projection onto the S (figure 5).
The symmetry line evaluated by means of the method
proposed in [5] is projected onto the sagittal plane and,
then, approximated by using a polynomial parametric
curve c = c(t) of degree 3 (figure 5b). The interpolation of
the points detected on the symmetry line is approximated
by using a special kind of parameterization, in such a
way that a linear map is defined between the value of
the parameter t and the position of the column segments
associated with the vertebrae. The value of t associated
with L5 is 0 and the value at the end of the column (T1)
is 1. The approximation of the points Pi is performed
by the least-squares method, for which the best coeffi-
cients are those which minimize the sum of the squared
Figure 4. The four characteristic parameters of a VDT virtual deviations of the curve from the points Pi :
working environment. 
 n
1 
S= Pi − c(ti )2 (3.1)
n i=1
When the individual’s anthropometric data and the
parameter c are known, it is possible to calculate the angle where the value of the parameter ti associated with the
β which measures the inclination between the femur and point Pi is evaluated as follows:
the base of the spine (L5) near the pelvis. The parame- i
ter β, through the pelvis configuration, greatly affects the j=1 Pj+1 − Pj
sitting posture [5]. ti = n−1 (3.2)
j=1 Pj+1 − Pj
The geometry of the back must be acquired so that
precise quantitative information can be taken. At this This parameterization makes it possible to normalize
purpose, the geometry of the back is acquired by using the symmetry line with little deviation from the real value,
a structured-light 3D scanner, which makes it possible regardless of the specific anthropometric characteristic of
to obtain the whole three-dimensional geometry of the the analysed subject.
back with a single scanning operation in a relatively short The spine is modeled as a multi-body kinematic struc-
time with a density sampling of about 0.75 mm. The ture comprising 17 motion segments (figure 6c). Each
32 L. DI ANGELO AND P. DI STEFANO

Figure 5. The point clouds acquired for the back of an individual, position markers and symmetry line.

motion segment of the lumbar spine is associated with as follows (figure 6 d):
a human vertebra and is defined as a rigid mechani-
ϕi = arccos[xi · xi−1 ] (3.3)
cal link. Thus the spine is modeled as an open kine-
matic chain in which each link is connected by ball-and- where xi is the direction of the (i-1)th link. The posi-
socket joints with both the previous and the subsequent tion and orientation of the end-effector of the i-th link
vertebrae. is expressed, according to the Denavit-Hartenberg con-
The dimension of each motion segment depends on vention, in terms of the joint variables by the following
the anthropometric characteristics which are chosen. The transformation matrix ii−1 T:
surveyed model of the spine, described by means of a ⎡ ⎤
polynomial function, is subdivided into 17 rectilinear cos(ϕi ) −sin(ϕi ) 0 ai
⎢ sin(ϕi ) cos(ϕi ) 0 0 ⎥
segments whose lengths correspond to the links of the i ⎢ ⎥
i−1 T = ⎣ 0 0 1 0⎦
(3.4)
kinematic model. For this purpose, 17 intervals (ti -ti+1
for i = 1, . . . .,18) of the parameter t of the curve c are 0 0 0 1
identified, which correspond to the location and length where ai is the length of the link projected along xi .
of 17 vertebrae. The angle ϕ i of the i-th link is evaluated Therefore the position of the end-effector of the i-th link

Figure 6. The multi-body kinematic structure modeling the spine.

 COMPUTER-AIDED DESIGN & APPLICATIONS 33

in homogeneous coordinates (Pi ), as a function of the commands are generated by the trained neural network
origin of the kinematic chain (P0 ), can be expressed as: and presented to Safework R
.
The neural network which has been used is of a back-
⎛ ⎞ propagation type and has been implemented in the MAT-
i Lab platform. It consists of 4 nodes in the input layer, 144
Pi = ⎝ j−1 T ⎠ · P0
j
(3.5) and 72 nodes in the two hidden layers and 17 nodes in the
j=1 output layer. The four nodes in the input layer are associ-
ated with the environment configuration parameters and
each node in the output layer provides the inclination
4. Software and tests
angle between two connected links (Figure 7). The activa-
The proposed method has been developed in a commer- tion functions used are sigmoid – tangent for the hidden
cial CAD environment (CATIA V5 R12), by using the layers and linear functions for the output layer.
Visual Basic Developer environment, which implements In order to be able to generalize the results obtained
the Safework R
virtual manikin (DASSAULT SYSTEMS, in the posture reconstruction of a specific individual to
2014). Safework R
provides a digital geometric represen- apply them to others characterized by different anthro-
tation of humans and permits the analysis of the interac- pometric parameters, the input parameters a and b
tions between the working place and the user. The skeletal (figure 4) are then normalized. Parameters a and b are
structure of the human model comprises 100 indepen- divided by the eye height:
dent elements and the complete model has 148 degrees
a b
of freedom, which are enough to ensure realistic joint ā = ; b̄ = (4.1)
eye height eye height
movement capability. The spine is modeled with 17 ele-
ments (5 lumbar vertebrae + 12 thoracic vertebrae) and In order to train the ANN, in the present study, the
the neck is defined as a single rigid piece. It is a fully spine geometry of three different individuals has been
articulated spine that can be configured by setting the acquired for each of the simulated working environ-
values of the angles between the 17 elements. The pos- ments, which are obtained by combining the parameters,
tures and movements of the Safework virtual manikin previously defined, in a fully factorial set of experiments.
can be simulated by assigning control commands follow- The anthropometric parameters are depicted in figure 8
ing a direct kinematic approach or an inverse kinematic and the results for each individual are reported in table 1.
approach. The direct-kinematic control commands make That is to say, the 45 working environment configura-
it possible to set the degree of freedom of each manikin tions (figure 9) are obtained with 5 different values for
component and assign its posture. Those direct control parameter θ , three values for parameters a and b, and

Figure 7. Neural network scheme.

34 L. DI ANGELO AND P. DI STEFANO

Figure 8. Anthropometric parameters used to characterize an individual.

Table 1. Anthropometric data that characterize the individuals

analyzed.
Individual Individual Individual
Anthropometric parameters No.1 [mm] No.2 [mm] No.3 [mm]
Acromion – radial length 340 320 330
Axilla height 1390 1290 1440
Chest breadth 320 380 400
Chest height 1350 1270 1360
Crotch height 830 800 870
Hip breadth, standing 360 360 440
Radial - stylion length 240 270 270
Sleeve outseam 620 90 600
Stature 1780 1700 1860
Waist breadth 280 340 360
Waist height 1120 1010 1100
Eye height 760 730 800
Hip breadth, sitting 380 390 430
Shoulder – elbow length 380 360 370
Sitting height 900 890 920
Buttock – knee length 570 550 590
Knee height, sitting 590 570 610

Figure 9. The 45 working environment configurations used in

the posture experiments.
assuming a fixed value for parameter c (fixed value for β).
In any case, the working environment for the experiment
has been defined according to the optimum situation for
EN ISO 9241 [7]. same subject, and average values are assumed to be the
Some experiments have been conducted to verify the inclination angles of the motion segments of the spine.
repeatability of the posture assumed when an individual The initial training of the neural network, which has
repeats the sitting process for a given environmental con- been carried out by using the postural data of subject
figuration. This preliminary test has confirmed a good no.1 (table 1), has required 2200 training cycles. The
repeatability of the sitting posture when the hand posi- training process is interrupted, though, for a training
tion is pinpointed with a marker on the table. Once the mean-square error, for all training data, less than 10−5 .
repeatability of that posture has been verified, each of Figure 10 shows the output error of the neural network
the subsequent postures is adopted three times by the for each of the 17 characteristic angles used to define the
 COMPUTER-AIDED DESIGN & APPLICATIONS 35

Figure 10. Neural network output angular error for each of the 17 characteristic links used to define the spine geometry, for three
environment configurations of the same working place.

spine geometry, and for three environmental configura- The postures predicted by using the proposed method
tions. The maximum error does not exceed 2.4% of the are compared with the postures generated by the Safe-
inclination angle dimension. work inverse kinematic approach. The inverse kinematic
Once the input parameters (θ , a, b, c) are provided approach is supported by five types of postures (stand, sit,
to the system, the virtual environment is set up and reach, span, kneel) and the specific configuration can be
the manikin, defined by its anthropometric parameters, defined by constraining the manikin in the virtual envi-
adopts the corresponding posture. The proposed method ronment at up to 20 points. However, Safework develop-
automatically configures the virtual manikin to assume ers do not supply any detailed information about the 7
the postural attitude of the specific individual that has criteria necessary to define postures (DASSAULT SYS-
been previously analysed. TEMS [4]). Very little, if not anything, is known about

Figure 11. Comparison of the postures predicted with Safework (a) and with the proposed method (b) for a given working environment:
θ = 0°, a = 230 mm e b = 250 mm.
36 L. DI ANGELO AND P. DI STEFANO

Figure 12. Examples of predicted postures for three different environment configurations.

those criteria of use. Figure 11a illustrates an example individuals for whom the neural network has not been
of a Safework manikin posture in a pre-defined working trained. In fact, the postural attitude is not merely a func-
environment where the view direction and the position tion of the anthropometric parameters and can, rather,
of the pelvis, feet and hands are constrained in the vir- vary a great deal from one individual to another. Future
tual environment. Figure 11b, on the other hand, shows works should be addressed to define apposite parameters
the posture of the manikin for the same environment but suited to classify and describe classes of postural attitude,
this time simulated by the proposed method. The pos- such as koilorachic and kurtorachic attitudes.
tures predicted with the proposed method appear to be This approach can be extended, in the future, to imple-
more natural, since it faithfully reproduces the postural ment data related to new critical human activities in
attitude of the individual. The postures adopted for the which women and men could take non correct pos-
different configurations of the working environment are tures for long time due to new habits. It could be the
shown in figure 12. All these postures are assumed with case in young people which use intensively tablets and
environment parameters which fall outside the training mobiles.
range. The method could be improved by considering the
rotation of the vertebrae around the sagittal and coronal
5. Conclusion axes. In this way, it would be possible to simulate and pre-
dict those postures assumed, consciously or forcedly, by
The use of Artificial Neural Network to predict posture is workers in a typical factory workplace which are highly
not new in absolute, but new is the use of this technique to asymmetric.
simulate posture using information concerning the spine
configuration. With the proposed method it is possible
to predict the spine posture of a virtual human model ORCiD
that interacts with a virtual working environment. The
Luca Di Angelo http://orcid.org/0000-0002-5341-0500
model can implement the postural attitude of a reference Paolo Di Stefano http://orcid.org/0000-0001-5003-2084
individual in order to predict postures in specific settings
where a particular activity is carried on. The EDHM here
proposed overcomes the weakness of the typical methods
used in commercial DHM to predict postures, perform-
References
ing a prediction of a more natural posture. The method is [1] Apostolico, A., Cappetti, N., D’Oria, C., Naddeo, A., Ses-
devoted to the reproduction of the posture of the upper tri, M.: Postural comfort evaluation: experimental iden-
tification of Range of Rest Posture for human articular
trunk of VDT operators; this is a part whose posture is joints, International Journal on Interactive Design and
particularly difficult and complex to simulate. For these Manufacturing, 8(2), 2013, 109–120. http://dx.doi.org/10.
reasons, what has been verified in this work proves that it 1007/s12008-013-0186-z
can used in posture prediction for the whole body and in [2] Bataineh, M.; Marler, T.; Abdel-Malek, K.: Artificial Neu-
other working activities. ral Network-Based Prediction of Human Posture. Lecture
Notes in Computer Science, Digital Human Modeling
The tests of the implemented method show high accu-
and Applications in Health, Safety, Ergonomics, and Risk
racy in posture reproduction for a single individual. Management, Human Body Modeling and Ergonomics,
Some problems have been encountered, though, when Duffy, VincentG, Springer Berlin Heidelberg, 2013.
the method used to predict posture is extended to other http://dx.doi.org/10.1007/978-3-642-39182-8_36
 COMPUTER-AIDED DESIGN & APPLICATIONS 37

[3] Colombo, G.; Regazzoni, D.; Rizzi, C.: Ergonomic design Design and Applications, 11(1), 2014, 10–19. http://dx.
through virtual Humans, Computer-Aided Design and doi.org/10.1080/16864360.2013.834130
Applications, 10(5), 2013, 745–755. http://dx.doi.org/10. [13] Rezzoug, N.; Gorce, P.: Prediction of Fingers Posture
3722/cadaps.2013.745–755 Using Artificial Neural Networks, Journal of Biomechan-
[4] DASSAULT SYSTEMS, Human Builder: User manual, ics, 41(12), 2008, 2743–2749. http://dx.doi.org/10.1016/j.
CATIA V5 R12, 2014. jbiomech.2008.06.005
[5] Dellerman, N. J.; Haslegrave, C. M.; Chaffin, D.B.: Work- [14] Tolani, D.; Goswami, A.; Badler, N.: Real-time Inverse
ing postures and movements, Tools for evaluation and Kinematics Techniques for Anthropomorphic Limbs,
engineering, CRC Press, 2004. Graphical Models, 62(5), 2000, 353–388. http://dx.doi.
[6] Di Angelo, L.; Di Stefano, P.; Spezzaneve A.: A method org/10.1006/gmod.2000.0528
for 3d symmetry line detection in asymmetric postures, [15] Vink, P.; Hallbeck, S.: Editorial: comfort and discomfort
Computer Methods in Biomechanics and Biomedical studies demonstrate the need for a new model, Applied
Engineering, 16(11), 2013, 1213–1220. http://dx.doi.org/ Ergonomics, 43, 2012, 271–276. http://dx.doi.org/10.
10.1080/10255842.2012.659245 1016/j.apergo.2011.06.001
[7] ISO 9241-5: 1998. Ergonomic requirements for office [16] Volpe, Y.; Governi, L.; Furferi, R.: A computational model
work with visual display terminals. for early assessment of padded furniture comfort perfor-
[8] Jung, E S.; Choe, J.: Human reach posture predic- mance, Human Factors and Ergonomics in Manufactur-
tion based on psychophysical discomfort, International ing, 25(1), 2015, 90–105.
Journal of Industrial Ergonomics, 18, 1996, 173–179. [17] VSR. Optimization Based Kinematics Models for Human
http://dx.doi.org/10.1016/0169-8141(95)00080-1 Posture. Technical Report N. VSR – 04.11, 2003, Virtual
[9] Jung, E. S.; Kee, D.; Chung, M. K.: Upper Body Reach Soldier Research (VSR) Program, Center of Computer-
Posture Prediction for Ergonomics Evaluation Mod- Aided Design, The University of Iowa.
els, International Journal of Industrial Ergonomics, 16, [18] Wang, X. G.: A behaviour-based inverse kinemat-
1994, 95–107. http://dx.doi.org/10.1016/0169-8141(94) ics algorithm to predict arm prehension postures for
00088-K computer-aided ergonomic evaluation, Journal of
[10] Kee, D.; Jung, E. S.; Chang, S.: A man-machine inter- Biomechanics, 32, 1999, 453–460. http://dx.doi.org/10.
face model for ergonomic design, Computers & Indus- 1016/S0021-9290(99)00023-8
trial Engineering, 27, 1994, 365–368. http://dx.doi.org/10. [19] Yang, J.; Abdel-Malek, K.: Real-Time Inverse Kinemat-
1016/0360-8352(94)90310-7 ics for Humans, Proceedings of 2002 ASME Design
[11] Moes, N.C.: Digital human models: an overview of Engineering Technical Conferences. DETC2002/MECH-
development and applications in product and workplace 34239, September 29-October 2, Montreal, Canada.
design. In: Proceedings of TMCE (Tools and Methods [20] Zhang, B.; Horváth, I.; Molenbroek, J.F.M.; Snijders,
of Competitive Engineering) International Conference, C.: Using Artificial Neural Networks for Human Body
Ancona (2010). Posture Prediction, International Journal of Industrial
[12] Regazzoni, D.; Rizzi, C.: Digital human models and virtual Ergonomics, 40(4), 2010, 414–424. http://dx.doi.org/10.
ergonomics to improve maintainability, Computer-Aided 1016/j.ergon.2010.02.003