Eurographics/ IEEE-VGTC Symposium on Visualization 2010
G. Melançon, T. Munzner, and D. Weiskopf
(Guest Editors)
Volume 29 (2010), Number 3
Visual Analysis of Multi-Joint Kinematic Data
Peter R. Krekel1,2, Edward R. Valstar1,5, Jurriaan de Groot3, Frits H. Post2, Rob G. H. H. Nelissen1, Charl P. Botha2,4
1 Biomechanics and Imaging Group, Leiden University Medical Center
2 Computer Graphics, Delft University of Technology
3 Department of Rehabilitation Medicine, Leiden University Medical Center
4 Divison of Image Processing, Leiden University Medical Center
5 Biomechanical Engineering, Delft University of Technology
Abstract
Kinematics is the analysis of motions without regarding forces or inertial effects, with the purpose of understanding
joint behaviour. Kinematic data of linked joints, for example the upper extremity, i.e. the shoulder and arm joints,
contains many related degrees of freedom that complicate numerical analysis. Visualisation techniques enhance
the analysis process, thus improving the effectiveness of kinematic experiments.
This paper describes a new visualisation system specifically designed for the analysis of multi-joint kinematic data
of the upper extremity. The challenge inherent in the data is that the upper extremity is comprised of five cooper-
ating joints with a total of fifteen degrees of freedom. The range of motion may be affected by subtle deficiencies
of individual joints that are difficult to pinpoint. To highlight these subtleties our approach combines interactive
filtering and multiple visualisation techniques.
Our system is further differentiated by the fact that it integrates simultaneous acquisition and visual analysis of
biokinematic data. Also, to facilitate complex queries, we have designed a visual query interface with visualisation
and interaction elements that are based on the domain-specific anatomical representation of the data. The combi-
nation of these techniques form an effective approach specifically tailored for the investigation and comparison of
large collections of kinematic data. This claim is supported by an evaluation experiment where the technique was
used to inspect the kinematics of the left and right arm of a patient with a healed proximal humerus fracture, i.e.
a healed shoulder fracture.
1. Introduction
Kinematic data describes the movement of limbs and is used
in biology, sports, orthopaedics and rehabilitation medicine.
The data is generally acquired using motion tracking sys-
tems, imaging systems or computer simulation. Examples of
motion tracking systems are Optotrak (Northern Digital Inc.,
Waterloo, Canada), which uses optical sensors, and Flock
of Birds (Ascension Technology Coorporation, Burlington,
USA), which uses electromagnetic sensors. The acquired
kinematic data is used to monitor surgical interventions or
to help answer fundamental research questions on kinematic
behaviour.
Despite the widespread use of kinematic analysis method-
ologies, creating visual representations of motion data that
support clinically relevant conclusions is challenging. The
most common method for depicting kinematic output is the
angle-angle plot. This is a two-dimensional plot that displays
how a certain joint angle relates to another joint angle. See
Figure 1 for an example of a series of standard angle-angle
plots. Angle-angle plots are limited to depicting two param-
eters, even though joint kinematics are often correlated in
three or more dimensions. For complex research questions
this may result in a large number of angle-angle plots. For
example, one publication by De Groot et al. includes a total
of 27 angle-angle plots [De 97]. In our opinion these plots
are functional when exact numerical values and relations are
required. However, for the exploration of kinematic data al-
ternative representations may be more informative.
This inspired us to create a system for the analysis of com-
c© 2010 The Author(s)
Journal compilation c© 2010 The Eurographics Association and Blackwell Publishing Ltd.
Published by Blackwell Publishing, 9600 Garsington Road, Oxford OX4 2DQ, UK and
350 Main Street, Malden, MA 02148, USA.

Krekel et al. / Visual Analysis of Multi-Joint Kinematic Data
CLAVICLE
0 100 200
−50
0
50
100
ax
ia
l r
ot
at
io
n
abduction
SCAPULA
0 100 200
−50
0
50
sp
in
al
ti
lt
abduction
HUMERUS
0 100 200
−200
−100
0
100
200
ax
ia
l r
ot
at
io
n
abduction
Figure 1: An example of three angle-angle plots. These plots
show the relation between elevation of the arm (abduction)
and other angles of three joints of nine subjects. Image cour-
tesy of Frans Steenbrink.
plex, multi-joint kinematic data that gives insight in relation-
ships between joint angles that would otherwise require a
predetermined hypothesis. In order to assist this process, we
have employed both forward and inverse visual query tech-
niques in our framework. With the former, researchers can
inspect the range of motion of multiple joints and find the
relationships between the available DOFs. With the latter,
the joint configurations that were used to reach a queried
location are extracted, functionality that is useful in investi-
gating for example compensatory kinematics in pathological
joints. We believe that this system could eventually lead to
new observations and different focus with respect to kine-
matic coupling of degrees of freedom (DOFs).
The contribution of this work is a comprehensive new ap-
proach to the visual analysis of complex multi-joint kine-
matic data. To the best of our knowledge, visual analysis
techniques have not yet been proposed for this type of kine-
matic data. Novel characteristics of this approach include the
following:
• Our system integrates real-time visualisation and acquisi-
tion. In other words, the visualisation process starts during
data acquisition, enabling the operator to steer the acquisi-
tion process, guided by conclusions drawn from the visual
analysis.
• Our forward visual query interface combines interaction
and visual feedback in an integrated anatomical repre-
sentation, allowing users to perform complex queries in
a recognisable and therefore straightforward manner.
• We demonstrate the utility of our work on the kinematic
data acquired of a proximal humerus fracture patient.
Supporting our main technical contributions,
the complete implementation of this approach is
available as open source at the following URL:
http://fobvis.googlecode.com.
The remainder of this article is structured as follows: in
Section 2 we discuss existing literature on kinematic data
visualisation. In the subsequent section we describe our vi-
sualisation framework, including the kinematic model and
filtering mechanisms that we apply. In Section 4 we describe
an evaluation experiment. For this experiment we recorded
the motion patterns of a subject with a healed shoulder frac-
ture and demonstrate that the visualisation framework en-
ables researchers to analyse the recorded motion patterns in
great detail. Lastly, we discuss the contributions and limita-
tions of our system and conclude with a prospect on future
work.
2. Related Work
Much of the research on kinematic analysis finds its origins
in gait analysis, the study of locomotion [Whi06]. Many im-
provements to the standard angle-angle plot originate from
this field. For example, by adding a third dimension to the
plots, an additional parameter can be visualised [MCHS05].
In addition, by color coding the graphs, another parameter
can be added, resulting in a total of four parameters of a mo-
tion recording that can be visualised [MS04]. The drawback
of these approaches is that the straight-forward addition of
the third dimension unnecessarily complicates the interpre-
tation of the data. Also, using colour to represent a continu-
ous parameter is ill-advised, especially if this channel could
later be useful for example to distinguish between patient
measurements, a categorical parameter [Mac86].
The work most related to our research is that of Keefe et
al. [KERC09]. Their visualisation system is an excellent ex-
ample of how a multiple-view approach effectively shows
relationships within kinematic data. Using small multiples
they visualise cycles of motion of pig jaws during eating.
Although many parallels exist between their and our work,
their visualisation technique is specifically targeting sequen-
tial data, whereas in our data we are interested in visualising
the relationships between multiple connected joints. Simi-
larly, Chen et al. used the approach of small multiples for
the cyclic patterns of bat wings during flight [CFSL07]. Al-
though they use many markers on the bats’ wings, there is
no joint decomposition with subsequent analysis of a multi-
joint kinematic model.
With regards to range of motion visualisation for the up-
per extremity there are several examples of applied visuali-
sation techniques. A basic approach is presented by Ct et al.,
who use 2-D projected stick figures to show the kinematic
results of a hammering task [CRM∗05]. This visualisation
suffices when looking at joint height, but does not reveal the
kinematic relationships.
In previous work we presented a technique for visualis-
ing range of motion of the shoulder joint [KBV∗06]. The
described pre-operative planning system visualises the sim-
ulated range of motion of the glenohumeral joint with a
moveable prosthesis. Although the comparative visualisa-
tion techniques are effective for the glenohumeral joint,
these techniques do not hold for the analysis of a multi-joint
kinematic chain. The main reason for this is that most joints
do not function as ball-and-socket joints.
c© 2010 The Author(s)
Journal compilation c© 2010 The Eurographics Association and Blackwell Publishing Ltd.

Krekel et al. / Visual Analysis of Multi-Joint Kinematic Data
Van Sint Jan et al. presented an interesting system that vi-
sualises the kinematics of multi-joints [VCR98] . Their work
uses computer tomography along with kinematic recordings
to link bone morphology of the fingers to kinematics. Their
visualisation method is limited to 3-D playback of the kine-
matics using patient-specific surface models.
Analogue to path planning techniques in robotics, a multi-
joint chain can be described in configuration space [BL91,
LP90]. The term configuration refers to a single pose of the
chain of joints. The individual joints have a local range of
motion that determine the total set of possible configura-
tions, i.e. the configuration space. Our system is built around
this concept, with separate views and filtering mechanisms
for the DOFs of the joints and for the total range of motion
of the limb.
In literature, similar visualisation approaches exist, for
example Abdel-Malek et al. [AMYBT04] and Lenarcˇicˇ et
al. [LK06], who describe multi-joint kinematic models with
accompanying visualisations of the configuration space or
reachable arm space. An interesting supplement to these ref-
erences dates back to 1955, where a similar range of motion
visualisation technique was used to design aeroplane cock-
pits [Dem55]. Although these visualisations give insight in
the reachable arm space, they do not disclose the underlying
kinematic dependencies. To our knowledge, no technique
exists that visualises both the DOFs of a kinematic model
and the resulting functional range of motion.
3. Methods
3.1. Requirements Analysis
To catalogue the requirements of an improved approach to
visualising range of motion measurements, we used the Del-
phi method [RW99]. Two human movement scientists and
four orthopaedic surgeons of different clinical institutions,
reflecting our target audience, were questioned using a list
of propositions and a number of example visualisations. The
complexity of the propositions varied, ranging from ‘Quan-
tifying measurements is more important than visualising
them’ to propositions as ‘I can use this example visualisa-
tion to track the progression of a muscular deficiency’.
Important conclusions that followed from this question-
naire were the following:
• Clinicians prefer more intuitive visualisations, whereas
human movement scientists prefer visualisations that give
access to more quantitative information, regardless of the
additional clutter that comes with this information. All of
the clinicians indicated that they were willing to use only
the most simple visualisation in their conversations with
patients, fearing that any visualisation other than a simple
shoulder picture would be too difficult to understand for
the average patient. To accommodate this requirement, a
clear distinction is made throughout this work between vi-
Figure 2: Visualisation concepts that were presented to the
participants of the questionnaire. Participants were asked to
rate several aspects with regards to clarity and usefulness
before and after an explanation was given. See Section 3.1
for a description of the subfigures.
sualising a subject’s function and visualising the range of
motion of individual joints.
• With respect to the possible benefits of a new visual
analysis system, participants indicated that assessment of
pathology, follow-up of patients and communication with
colleagues were of foremost importance.
• Due to time constraints, clinicians would be willing to
spend only a few minutes analysing the kinematic data.
This introduces the requirement that a kinematic visuali-
sation technique has to be fast and intuitive.
• The reachable arm space visualisation by Lenarcˇicˇ et al.
(see Section 2), used as an example visualisation, was
thought to be very useful. A similar visualisation disclos-
ing more details on the separate joints was expected to
fulfil most of the requirements.
Figure 2 shows some of the visualisation concepts pre-
sented to the participants of our questionnaire. The proposed
visualisations were the following (see figure):
a. An integration of the DOF values with the animated
bone model representation. We have implemented this
technique and discovered that occlusion and continu-
ously changing coordinate systems make this visualisa-
tion counterintuitive.
b. A schematic 2-D plot of the various parameters. This con-
cept was eventually extended and implemented in our 3-D
Pose View (see Section 3.5.2).
c. Integration of the parallel coordinates plot with the spa-
tial location of the joints. Although this would make the
semantics of the plot more intuitive, it was expected that
the anatomical location would complicate the visual rep-
resentation and hence understanding of relations between
angles.
d. A segment visualisation with various types of endpoints
to depict parameters. This view merges the visualisation
c© 2010 The Author(s)
Journal compilation c© 2010 The Eurographics Association and Blackwell Publishing Ltd.

Krekel et al. / Visual Analysis of Multi-Joint Kinematic Data
Figure 3: The FobVis motion tracking software. Sensors are
attached to the skin and depicted by red spheres. Subse-
quently, bony landmarks are registered relative to these sen-
sors and depicted by yellow spheres. The surface models are
rigidly transformed in accordance with the sensors and their
respective bony landmarks.
of individual joint angles and the total functional range
of motion. We consider this undesirable because each an-
swers a different set of research questions.
e. Volume visualisation with a slice-viewer and visualisa-
tion method to depict trajectories. For this specific appli-
cation, volume rendering would not contribute any spe-
cific advantages. Furthermore, the conversion to volumes
would unnecessarily complicate the interactivity of our
system.
f. A spherical representation of the joint, mapped to a 2-D
plot. Not all joints are spherical joints, making it hard to
defend this visual encoding. In addition, this representa-
tion only allows for two degrees of freedom (DOFs) per
joint.
Although these visualisation techniques were not used,
various aspects were extracted and incorporated in the final
visualisation system described herein.
3.2. Software
The software used to record shoulder motion is FobVis, a
package developed by our institution. FobVis is currently
built for the Flock of Birds electromagnetic system, but the
generic design facilitates the use of other motion tracking
systems, for example Optotrak. It has been implemented as
a state machine, the transitions between states reflecting the
motion tracking procedure described below. Figure 3 shows
a screenshot of FobVis.
The work described in this article is implemented as a
module in the FobVis software and directly uses the recorded
data as input for kinematic analysis. Both the FobVis soft-
ware and the visual analysis module are available as open
source.
3.3. Motion Tracking
Motion tracking was performed using the Flock of Birds mo-
tion tracking system. The workflow for recording motion is
in accordance with Kontaxis et al. [KCJV09] and consists of
the following steps:
1. Sensors are attached to the body of the subject.
2. The positions of prominent bony landmarks relative to
the sensors are registered using a motion tracked pointing
device.
3. The subject follows the movement instructions of the re-
searcher, during which the positions and rotations of the
sensors are recorded. In combination with the bony land-
mark positions relative to the sensors this gives sufficient
information to track motion of the bones, with a small
error due to the sensors being attached to the skin rather
than to the bones.
In common motion recording protocols, the subject is in-
structed to make specific movements that are expected to an-
swer the research questions under investigation. Our tech-
nique is based on the principle that as much data should
be collected as possible. The acquisition process is closely
monitored by the operator, assisted by the real-time visual-
isations of our system. After the acquisition, the investiga-
tor can filter data and focus on the specific type of move-
ments he would like to see. The advantages of this approach
are that recording motion is not restricted to specific move-
ments, making the recording procedure less error-prone. In
addition, the investigator can pose additional research ques-
tions after doing the measurements, as a large collection of
motion data is included in the visualisation.
The motion recording system is continuously updated at
25 frames per second, giving immediate feedback to the
researcher. This allows the researcher to determine when
enough data has been gathered by inspecting the visualisa-
tions.
3.4. Kinematic Model
Kinematic models of the human body usually consist of a hi-
erarchical structure of kinematic chains. A kinematic chain
is a series of linked rigid body segments connected by joints
with one or more rotational degrees of freedom. The motion
of a kinematic chain is defined by the link lengths and the
variation of joint angles. The lengths are assumed to be con-
stant for a given individual, so the postures and motions can
be completely described by the joint angles.
Different kinematic models can be used to analyse motion
data. We have defined our kinematic model in accordance
with the authoritative work on upper extremity kinematics
c© 2010 The Author(s)
Journal compilation c© 2010 The Eurographics Association and Blackwell Publishing Ltd.

Krekel et al. / Visual Analysis of Multi-Joint Kinematic Data
Figure 4: Kinematic model as used within our system. The
chain of rotations along various axes begins at the spine
and terminates at the hand. The character g refers to the
global coordinate system; t, c, s, h and f refer to the tho-
rax, clavicle, scapula, humerus and forearm coordinate sys-
tem respectively. The model is completely described in Wu et
al. [WvV∗05].
by Wu et al. [WvV∗05]. The assembly and connectivity of
the modeled joints is depicted in Figure 4, rotations starting
at the first degree of freedom of the spine and terminating
at the last degree of freedom of the elbow. Quantified angles
are defined relative to the proximal (preceding) joint as well
as relative to the global coordinate system. These angles are
calculated in real-time as motion data is acquired.
3.5. Visualisation and Filtering
Multiple linked views are used to analyse the kinematic data
(see Figure 5). The degrees of freedom View, or DOF View,
shows the local range of motion of each of the DOFs of the
individual joints. The Pose View is used to depict the total
functional range of motion of the multi-joint configuration,
as well as spline curves to see a time window around poses.
A parallel coordinates plot is used to quantify relationships
between the parameters of the kinematic model. In addition,
an unlimited number of 2D plots of data values over time can
be added. Lastly, scatter plots can be generated. The different
views are discussed below.
During acquisition all views are continuously updated,
providing interactive feedback on the motion data acquired.
Large differences in range of motion are noticable, but gen-
erally and especially when working with large datasets, the
number of visible poses is too large to disclose valuable in-
formation. Visualisation and filtering form an integrated sys-
tem that allows the user to analyse and compare large collec-
tions of complex kinematic data.
To find interesting characteristics of the data the re-
searcher may want to omit data outside of a given range of
a specific DOF, inspect recordings that go through a point in
space or select a certain time range of the recordings where
something occurred that he found interesting. For this pur-
pose a number of filters were implemented. Filters can be
activated or deactivated, depending on the requirements of
the intended task. The user interface components of each of
the filters are integrated in the visualisation of the kinematic
aspect the filter acts upon.
Two motion recordings can be loaded simultaneously, al-
lowing for comparison of datasets. Examples are pre- and
post-operative measurements, left and right shoulders or a
(bundled) group of patients suffering from the same pathol-
ogy, compared to an equally large group of healthy subjects.
Motion recordings are assigned different colors and adjust
their alpha blending in each of the views in accordance with
the number of poses that are visible to optimize the amount
of information shown.
In the following subsections, we discuss each of the views
of our application, first focusing on the chosen visual repre-
sentation and then detailing the filtering possibilities for that
view.
3.5.1. DOF View
To determine how datasets are different from one another, re-
searchers will generally be interested in how the DOFs vary
in relationship to other DOFs. In the DOF View these range
of motion intervals are depicted in the form of joint wid-
gets (see Figure 6). Joint widgets can be added by selecting
a joint node and indicating which kinematic parameter is of
interest via a popup menu. The widget shows the minimum
and maximum value of the selected parameter for each of
the active datasets. To prevent clutter, joint widgets can be
collapsed by selecting their centerpoint.
Besides visualising the range of motion of a DOF, joint
widgets also function as the user interface element of DOF
filters. These are used to hide kinematic data in the linked
views, filtering the data based on a selected range of the
concerning DOF. The joint widget contains an orange pie-
shaped figure, its adjustable size modifying the filtered
range. Interaction with the widget updates the linked views,
showing only the kinematic data that passed all of the DOF
filters.
3.5.2. Pose View
In the Pose View the recorded poses are displayed as simple
line drawings (see topright of Figure 5). Joints can be dis-
c© 2010 The Author(s)
Journal compilation c© 2010 The Eurographics Association and Blackwell Publishing Ltd.

Krekel et al. / Visual Analysis of Multi-Joint Kinematic Data
Figure 5: A screenshot of the system. The DOF View visualises the individual DOFs of the different joints on demand. The
Pose View shows the recorded poses and thus visualises the functional range of motion. The parallel coodinates plot visualises
the interrelationships between DOFs. At the bottom of the interface 2D plots of data values over time can be added. Lastly, a
number of scatter plots can be in visualised in a separate frame.
Figure 6: The DOF View. Initially this 3D view only displays
the silhouette of a human torso with a schematic represen-
tation of the predefined connected joints. By selecting joints
the user can add visual representations of the range of mo-
tion of the DOFs as defined by the applied kinematic model.
For each of these DOFs a joint widget is added that includes
a blue and a yellow bar representing the range of motion of
two different datasets. The orange pie-shaped parts are DOF
filters, used to display only a part of the data in the linked
views. To prevent clutter joint widgets can be collapsed by
selecting their centerpoint.
Figure 7: Spline curve visualisation. This visualisation can
be used to inspect a time window around a specific pose. An
advantage of this visualisation method is that it shows how
a subject reached for a specific area.
abled, transforming the line drawings to take into account
the disabled joint and its corresponding DOFs. This allows
researchers to analyse what part of the functional range of
motion can be ascribed to the range of motion of specific
joints.
Axial rotation is an import kinematic parameter as it is
often jeopardised in case of a pathology. Because each pose
c© 2010 The Author(s)
Journal compilation c© 2010 The Eurographics Association and Blackwell Publishing Ltd.

Krekel et al. / Visual Analysis of Multi-Joint Kinematic Data
(a)
(b)
(c)
Figure 8: Axial rotation visualisation. (a) Segments that
consist of a single line, for example the upper arm, have ax-
ial rotation that is not visible using the standard line draw-
ing. (b, c) Optionally these segments can be replaced by
corkscrew lines, spiraling inward or outward depending on
the sign and magnitude of the axial rotation. See Figure 9
for an example of this visualisation.
is represented by a collection of simple lines, the basic vi-
sualisation is not capable of visualising the axial rotation of
a segment. Optionally, the researcher can visualise axial ro-
tation by replacing the line segments with corkscrew repre-
sentations that rotate either inward or outward, depending on
the sign and magnitude of the axial rotation (see Figure 8).
Alternatively, users can choose to visualise the poses as
interpolated spline curves. Spline curves originate from a se-
lectable set of joints, their length depending on the size of the
adjustable time window. The benefit of spline curves is that
they are time dependent and can therefore be used to anal-
yse the time window around a specific pose. A disadvantage
of this visualisation technique is that the view becomes clut-
tered when using large time windows for large quantities of
data.
In the Pose View a skeleton surface model represents an
individual pose when required. This includes the visualisa-
tion of newly acquired poses during recording and poses that
have been selected in the parallel coordinates plot.
The Pose View includes a pose filter that uses the position
of the hand of the skeleton surface model. By dragging the
hand to different positions, recorded poses that do not come
within a scalable sphere around the hand are occluded. In
this way the filter can be used to visualise functional infor-
mation, showing how a subject reached for a specific area.
See Figure 9 for an example of this filter.
While the hand is dragged to different locations the skele-
ton surface model snaps to the closest pose that passes the
filter. If none of the recorded poses pass the filter, a simple
inverse kinematics model is used to determine the arm posi-
tion for the new location of the hand. The inverse kinematics
model determines the gradient of each of the DOFs and ap-
plies weighted rotations in accordance with these gradients.
The individual DOFs are limited to the range of motion in-
Figure 9: The Pose View filter. The hand can be dragged to
a position in space. This position is then used for an inverse
query to determine how a subject reached for that position.
Note that the subject of the yellow dataset reached for the
point in a different manner compared to the subject of the
blue dataset. Also notice that the blue lines of the upper arm
are showing a larger amplitude, indicating that the axial ro-
tation of the upper arm of this subject had a greater magni-
tude.
terval as determined from the motion recordings. The inverse
kinematics model is only used for realtime visual feedback
during interaction.
3.5.3. Parallel Coordinates Plot
The parallel coordinates plot serves as a quantitative confir-
mation tool for motion patterns found in the DOF View and
Pose View. Each recorded pose is represented by a single
spline (see Figure 10). The view can be configured to ac-
comodate the researcher’s requirements, plotting any of the
parameters in sequence. The selection and ordering of pa-
rameters depends on the research question and aspects of
interest found in the DOF View or Pose View.
We adopted curved (cardinal) splines to distinguish mul-
tiple splines going through the same values, as was proposed
by Graham et al. [GK03]. In combination with alpha blend-
ing this enables us to display a large number of poses without
losing the focus on relationships between multiple DOFs of
the kinematic data. Optionally, the user can switch to linear
splines, as these may reveal linear relationships that are not
visible when using cardinal splines.
The parallel coordinates view is linked to the other views.
Selection of a specific spline will update the Pose View to
show the selected pose. In turn, when any of the filters in the
DOF View or Pose View are modified, the parallel coordi-
nates view is updated.
3.5.4. 2D Plots
In addition to the above views, 2D plots of data values over
time can be added (see Figure 10). Besides the informative
c© 2010 The Author(s)
Journal compilation c© 2010 The Eurographics Association and Blackwell Publishing Ltd.

Krekel et al. / Visual Analysis of Multi-Joint Kinematic Data
Figure 10: Parallel coordinates plot for two sets of motion recordings. The plotted parameters can be customised. Below the
parallel coordinates plot are two 2D plots of a specific parameter, in this case global elevation, over time for each dataset.
aspect of these 2D plots, the researcher can use them as a
timeline filter. Highlighting a range in one of the 2D plots
causes the data outside of this range to be hidden. In practice
this filter is frequently used complementary to a DOF filter.
The latter filters a selection range of the DOFs, whereas the
timeline filter allows selection of one of the intervals that
passed the DOF filter.
3.5.5. Scatter Plots
Scatter plots can be generated by right-clicking between two
axes of the parallel coordinates plot. All scatter plots are con-
tinuously updated, thereby only showing the recordings that
have passed the active filters. Scatter plots are capable of re-
vealing more complex relationships than possible with the
parallel coordinates plot. The disadvantage of using scatter
plots is that they can only show two parameters at once. As
such, a scatter plot is comparable to angle-angle plots.
4. Evaluation
An evaluation experiment was performed with a patient who
suffered a proximal humerus fracture injury. This is a com-
plex fracture where the shoulder part of the upper arm shat-
ters into multiple parts, each connected to a tendon of the
different muscles of the rotator cuff, i.e. the muscles that pro-
vide shoulder stability. After the fracture occurred the shoul-
der was operated on to reconstruct the normal anatomy. The
patient was seen eight months after trauma. In this time pe-
riod the formerly fractured shoulder regained similar range
of motion as compared to the healthy opposite side.
To demonstrate how our system can be employed to anal-
yse the kinematics of these fractures we used the Flock of
Birds system and instructed the patient to perform multiple
elevation tasks in various manners. This included a crouched
and extended attitude, forward elevation (flexion) and side-
ways elevation (abduction). Subsequently, the resulting mo-
tion recordings were analysed using our system.
Figure 11: Approach to answering the research questions,
further explained in Section 4.
No literature is available describing the kinematic changes
for this specific type of injury. However, in normal shoulder
kinematics the scapula (the shoulder blade) moves in unison
with the humerus (the upper arm bone) during an elevation
task. This is called the scapulohumeral rhythm. In healthy
subjects this is a near linear relationship where for every de-
gree of elevation the scapula upward rotation increases with
±0.5 degrees [EMK05]. With this experiment we wanted
to assess whether this relationship still holds for the healed
shoulder.
In summary, the research questions were:
1. Is there an impairment, and if so, where is it located?
2. Has the scapulohumeral rhythm of the formerly fractured
side changed?
The steps followed to find the answers to the research
questions are depicted in Figure 11. The Pose View (Fig-
ure 11a) demonstrates that the range of motion of the healthy
side is larger than the range of motion of the formerly frac-
tured side. The difference in range of motion is also evident
from the DOF-view (Figure 11b, *-mark). Using the DOF
filters we adjust the visible elevation interval (marked **)
c© 2010 The Author(s)
Journal compilation c© 2010 The Eurographics Association and Blackwell Publishing Ltd.

Krekel et al. / Visual Analysis of Multi-Joint Kinematic Data
Figure 12: Parallel coordinates plot of the mirrored for-
merly fractured shoulder (yellow) and the healthy shoulder
(blue). The top subfigure shows the complete collection of
splines, each one representing a recorded pose. The remain-
ing subfigures show the same plot with different DOF filters
for the global elevation parameter. The filters are visible on
the left of the image. Notice the different values of scapula
protraction and tilt at different global elevation angles, in-
dicating that the scapula hardly compensates for the loss of
humeral range of motion.
and restrict our data to a specific elevation plane to allow for
comparison of the two recordings. Subsequently, the differ-
ent scapulohumeral rhytmn becomes visible, both by means
of the surface models and by means of spline curves. To
quantify the difference we inspect the parallel coordinates
plot seen in Figure 12.
From this analysis we conclude that there is indeed an
impairment. Interestingly, the deficiency does not just find
its origins in the formerly fractured humerus, but also in the
mobility of the shoulder blade. A possible reason for this can
be the experience of pain or increased muscle tension.
Secondary to the above, the visualisation demonstrates
that the scapula of the formerly fractured side does not move
in unison with the humerus. Specifically, we conclude that
the scapulothoracic rhythm of the right shoulder deviates
from that of the left shoulder in that scapula protraction and
tilt lag behind of humeral elevation.
A limitation we encountered during this experiment is
that the limited accuracy of bony landmark registration may
result in an offset for the coordinate systems or segment
lengths. As was shown by Karduna et al. [KMMS01], this
inaccuracy does not prevent analysis of motion patterns
of individual datasets. However, when comparing multiple
datasets as we did in this evaluation experiment, inaccuracy
may lead to a relative offset in the coordinate systems used
for kinematic analysis.
An involved orthopaedic surgeon later stated that he was
impressed with the visualisation and speed of the system.
His feedback was valuable for the continuation of this re-
search.
5. Conclusions and Future Work
In this paper we have presented a novel visualisation system
for the analysis of kinematic recordings for the upper ex-
tremity in combination with a new method of data sampling.
The system currently supports six visualisation techniques
that are collectively used to filter motion data and inspect
relationships between the various DOFs. Although designed
for the upper extremity, the presented techniques can be ap-
plied to other multi-joint chains.
The benefit of our visualisation system is that users can
analyse kinematic recordings without predetermined hy-
potheses. It allows users to find interesting patterns that
could otherwise only be found through a large number of
angle-angle plots. Using the step-by-step approach described
in our evaluation experiment the majority of kinematic re-
search questions can be answered.
As was shown in the evaluation experiment, the visual
analysis technique is effective for comparison of two record-
ings, bearing in mind that inaccuracy of the motion record-
ings may lead to incorrect representations. We are aware that
these inaccuracies affect the kinematic analysis, but wish
to emphasise that this problem also holds for conventional
kinematic analysis. In addition, we have shown that our vi-
sualisation system is robust to these errors to a certain extent.
An interesting extension of the system would be to incor-
porate acquisition hints based on an automatic comparison
of the acquired data with a collection of kinematic measure-
ments. In this way the kinematic measurements themselves
can zoom in on interesting characteristics, thereby not only
relying on the assessments of the researcher.
An important message of this work is that kinematic be-
haviour requires a combination of visualisation and filter-
ing techniques, as was demonstrated with our system. The
modular design of our system allows for the implementation
of additional filters and visualisation methods and raises the
question whether this work should be taken one step further.
It is conceivable that modular approaches commonly seen in
image processing may be applicable for kinematic data, even
though image data and kinematic data are of a very different
nature. Future work includes the development of a data-flow
network editor where filtering and visualisation modules can
be connected to produce a specific kinematic visualisation.
Acknowledgements
This work was supported by a grant from the Dutch Arthri-
tis Association (Reumafonds). We acknowledge Professor
Roger Emery, Imperial College London, for his assistance
in this research and Mr Peter Reilly, Imperial College Lon-
don, for the formidable surgical result obtained.
c© 2010 The Author(s)
Journal compilation c© 2010 The Eurographics Association and Blackwell Publishing Ltd.

Krekel et al. / Visual Analysis of Multi-Joint Kinematic Data
References
[AMYBT04] ABDEL-MALEK K., YANG J., BRAND R., TAN-
BOUR E.: Towards understanding the workspace of human limbs.
Ergonomics 47, 13 (oktober 2004), 1386–405.
[BL91] BARRAQUAND J., LATOMBE J.-C.: Robot Motion Plan-
ning: A Distributed Representation Approach. The International
Journal of Robotics Research 10, 6 (december 1991), 628–649.
[CFSL07] CHEN J., FORSBERG A. S., SWARTZ S. M., LAID-
LAW D. H.: Interactive Multiple Scale Small Multiples. In Pro-
ceedings of IEEE Visualization (2007).
[CRM∗05] CÔTÉ J. N., RAYMOND D., MATHIEU P. A., FELD-
MAN A. G., LEVIN M. F.: Differences in multi-joint kine-
matic patterns of repetitive hammering in healthy, fatigued and
shoulder-injured individuals. Clinical biomechanics (Bristol,
Avon) 20, 6 (juli 2005), 581–90.
[De 97] DE GROOT J.: The variability of shoulder motions
recorded by means of palpation. Clinical Biomechanics 12, 7-
8 (1997), 461–472.
[Dem55] DEMPSTER W. T.: Space Requirements of the Seated
Operator, 1955.
[EMK05] EBAUGH D. D., MCCLURE P. W., KARDUNA A. R.:
Three-dimensional scapulothoracic motion during active and
passive arm elevation. Clinical Biomechanics 20, 7 (2005), 700–
709.
[GK03] GRAHAM M., KENNEDY J.: Using curves to enhance
parallel coordinate visualisations. Information Visualization,
2003. IV 2003. Proceedings. Seventh International Conference
on (2003).
[KBV∗06] KREKEL P. R., BOTHA C. P., VALSTAR E. R.,
DE BRUIN P. W., ROZING P. M., POST F. H.: Interactive
simulation and comparative visualisation of the bone-determined
range of motion of the human shoulder. In Proceedings of Sim-
ulation and Visualization (2006), Schulze T., Horton G., Preim
B., Schlechtweg S., (Eds.), SCS Publishing House Erlangen,
pp. 275–288.
[KCJV09] KONTAXIS A., CUTTI A. G., JOHNSON G. R.,
VEEGER H. E. J.: A framework for the definition of standard-
ized protocols for measuring upper-extremity kinematics. Clini-
cal biomechanics (Bristol, Avon) 24, 3 (2009), 246–53.
[KERC09] KEEFE D., EWERT M., RIBARSKY W., CHANG R.:
Interactive Coordinated Multiple-View Visualization of Biome-
chanical Motion Data. viscenter.uncc.edu 15, 6 (2009), 1383–
1390.
[KMMS01] KARDUNA A. R., MCCLURE P. W., MICHENER
L. A., SENNETT B.: Dynamic measurements of three-
dimensional scapular kinematics: a validation study. Journal of
biomechanical engineering 123, 2 (april 2001), 184–90.
[LK06] LENAR ˇCI ˇC J., KLOP ˇCAR N.: Positional kinematics of
humanoid arms. Robotica 24, 1 (2006).
[LP90] LOZANO-PÉREZ T.: Spatial planning: a configuration
space approach. Autonomous robot vehicles (1990).
[Mac86] MACKINLAY J. D.: Automating the Design of Graphical
Presentations of Relational Information. ACM Transactions on
Graphics 5 (1986), 110–141.
[MCHS05] MANAL K., CHANG C.-C., HAMILL J., STANHOPE
S. J.: A three-dimensional data visualization technique for re-
porting movement pattern deviations. Journal of biomechanics
38, 11 (2005), 2151–6.
[MS04] MANAL K., STANHOPE S. J.: A novel method for dis-
playing gait and clinical movement analysis data. Gait & posture
20, 2 (2004), 222–6.
[RW99] ROWE G., WRIGHT G.: The Delphi technique as a fore-
casting tool: issues and analysis. International Journal of Fore-
casting 15, 4 (1999), 353–375.
[VCR98] VAN SINT JAN S. L., CLAPWORTHY G. J., ROOZE
M.: Visualization of Combined Motions in Human Joints. IEEE
Computer Graphics and Applications 18, 6 (1998).
[Whi06] WHITTLE M.: Gait analysis: an introduction, 4th ed.
Butterworth-Heinemann Publishing, Oxford, 2006.
[WvV∗05] WU G., VAN DER HELM F. C. T., VEEGER H. E. J.,
MAKHSOUS M., VAN ROY P., ANGLIN C., NAGELS J., KAR-
DUNA A. R., MCQUADE K., WANG X., AL. E.: ISB recom-
mendation on definitions of joint coordinate systems of various
joints for the reporting of human joint motion–Part II: shoulder,
elbow, wrist and hand. Journal of Biomech 38 (2005), 987–992.
c© 2010 The Author(s)
Journal compilation c© 2010 The Eurographics Association and Blackwell Publishing Ltd.