Anda di halaman 1dari 7

13th World Congress in Mechanism and Machine Science, Guanajuato, Mxico, 19-25 June, 2011

IMD-123

A new method to solve kinematic consistency problem based on optimization


techniques and Euler parameters
J. Ojeda*
University of Seville
Seville, Spain

J. Mayo
University of Seville
Seville, Spain

J. Martnez-Reina
University of Seville
Seville, Spain

displacement signals are differentiated to obtain the


velocities and accelerations required to solve the inverse
dynamic problem. This high-frequency noise is usually
reduced using different filtering techniques. But there is
another kind of errors which cannot be removed by
filtering because such perturbations typically contain the
same frequencies as those of the movement [3]. These
errors are caused by the skin motion artifact [4,5,6]; due
to skin movements, the markers displace and rotate as a
rigid body relative to the underlying bone. Due to the
skin motion artifact, the processed kinematic data do not
ensure that the kinematic constraints associated to the
biomechanical model are fulfilled, i.e., the kinematic data
are inconsistent with the biomechanical model. The
inverse dynamic analysis also requires that velocities and
accelerations are known. An usual method to obtain those
variables involves the use a polynomial interpolation of
the coordinates and its time derivatives. This procedure
does not ensure that the constraint velocity and
acceleration equations are fulfilled, even if the position
data are
kinematically consistent. Consequently,
spurious joints reaction forces and net moments-of-force,
associated to the constraint violations, are generated in
the solution of the inverse dynamic problem [7]. These
errors can also have significant effects on the estimation
of the lines of action and lever arms of the muscles and
the forces transmitted in muscles and other structures [8].
Efforts have been made to improve measurement
techniques to minimise skin movement artifacts [9] but
they cannot be eliminated unless markers are applied to
the bones directly or through bone-pins [10]. Therefore,
spatial reconstruction of the musculoskeletal system and
calculation of its kinematics and kinetics via a skin
marker based multi-link model should take account of
skin movement artifacts [8].
In this work, different methods for generating
kinematically consistent data are compared. The
proposed methods use the filtered data. This is the usual
scheme in which filtering and kinematic consistency
numerical schemes are consecutive steps. However, some
authors [11] propose the application of an integrated
smoothing-differentiation-projection approach. They
smooth and differentiate the kinematic signals in a single
step using the Newmark integration scheme and then
project the positions and the obtained smoothed
velocities and accelerations to the biomechanical model

AbstractThe estimation of the skeletal motion


obtained from marker-based motion capture systems
affects the results of the kinematic and dynamic analysis
of biomechanical systems. A new method based on
optimization techniques and Euler parameters has been
developed in this work and compared with other
approaches found in the literature. Results show that for
a theoretical motion this new method gives better results
with a lower level of error.
Keywords: Biomechanics, Kinematic consistency, Euler parameters

I Introduction
A typical problem in Biomechanics is the determination
of loads supported by joints or muscle forces necessary to
perform a particular movement. However, direct
measurement of these forces requires invasive techniques
whose implementation is difficult and rarely viable. To
circumvent this problem, inverse dynamics is used to
estimate the driving forces required to perform the
movement [1,2].
Since deformation of bones may be neglected for most
human motions, the skeleton can be modeled by the
multibody system method. The system consists of rigid
bodies connected by ideal frictionless joints. An array of
at least three markers per segment is needed for the
definition of the position and orientation of a rigid body
in space. With the mechanical model of the body and the
information of the kinematics, obtained from the markers
and reaction forces on the ground from the force
platforms, the dynamics of the skeletal system, described
by its equations of motions, can be inverted to determine
the driving forces in the joints. In this process it is
essential to start up with reliable measurements in order
to get accurate results. The estimation of the skeletal
motion obtained from marker-based motion capture
systems is known to be affected by significant errors.
One source of errors is the noise introduced by the
motion capture system. This noise 1is amplificated
because of its high frequency content when the raw

*
joaquinojeda@us.es

juana@us.es

jmreina@us.es
13th World Congress in Mechanism and Machine Science,
Guanajuato, Mxico, 19-25 June, 2011
1

13th World Congress in Mechanism and Machine Science, Guanajuato, Mxico, 19-25 June, 2011

IMD-123

The local optimization method, also known as the


segmental optimization method (SOM), looks at the
problem under a different angle. It does not produce
consistent data, but minimizes the inconsistency in
all the coordinates of the system, both dependent and
independent.
The method uses least squares optimization to
minimize the difference between the coordinates
obtained with the model and the experimental
marker trajectories. Poses (position and orientation)
of multi-link models are obtained by sequentially
calculating the separate pose of each segment,
without considering joint constraints [13].
The coordinates vector qi, for each time step, is
estimated from:
2
n
n 3
2
min m j = min (mjk (q i ) m jk ) (4)
j =1

j =1 k =1

where ni is the number of markers associated with


segment i (at least three), m an experimental marker,
and m the corresponding model marker.
In this work, cartesian coordinates have been used to
describe the system configuration. Euler parameters
have been selected to describe the segments
orientation.
r
q i = Oi (5)
pi

constraint manifold. Another important characteristic of


the methods used in this work is the model parameters.
The proposed methods adjust the model coordinates to fit
a kinematic model to experimental movement data, but
they consider classical kinematic joints with fixed
centers. Lengths of segments are calculated as the
average of the lengths measured during the capture, or by
the simple direct measurement between anatomical
points. Some authors adjust not only the model
coordinates but also joint parameters. Reinbolt et al. [12]
propose a two-level optimization approach that
simultaneously optimizes joint parameters and motion.
II. Methods to produce kinematic consistent data
using Cartesian coordinates
Different methods can be found in the literature. They
can be divided into: 1) procedures that minimize the
kinematic inconsistency and 2) procedures that verify
kinematic constraints equations. One method of each
category plus a third one which tries to solve the
limitation of the previous methods will be explained in
this section
A. Adaptation model
Silva and Ambrsio [7] proposed a simple
methodology to systematically ensure the kinematic
data consistency: the kinematic positions are
modified in order to fulfill the constraint equations.
Furthermore, the velocity and acceleration of the
system are obtained by using the velocity and
acceleration equations, respectively. The model
coordinates q can be divided into dependent, qd, and
independent, qi, coordinates.

where rOi = [xi yi z i ]T is the position vector of center

of gravity of segment i, and pi = [e0 e1 e2 e3 ]i the


vector of Euler parameters. Because of the use of
Euler parameters, some constraints must be imposed
to the minimization problem in order to get realistic
parameters. Thus, the problem would be raised, for
each time step and for each segment, according to
equation:
T

q = q d (1)

qi

These authors estimate the independent coordinates


from the measured markers motion. They use these
non-consistent independent coordinates to calculate
the dependent coordinates solving the nonlinear
position problem.
(q ) = 0 (2)
where is the system of nonlinear equations
including just kinematic constraints. Consistent
velocities q& and accelerations q&& are obtained by
solving the velocity and acceleration equations of the
multibody system:
&
(3)
q = 0

n
n
T
i
i
min mj = min (Ai ( pi )rOMj
+ rOi mj ) (Ai ( pi )rOMj
+ rOi mj )

j=1

j=1
(6)
s.t.

1 ei 1 i [0,3]
3

2
i

=1

i=0

where Ai(pi) is the rotation matrix of segment i; and


riOMj the position vector in local coordinates of marker
j that we assume constant and known from anatomical
data.
C. Global optimization method
This method takes the best characteristics of the
previous two and solves their limitations. It does not
use inconsistent motion histories to drive the system,
but it completely ensures that the kinematic data are
consistent. The method is based on the search of an
optimal pose of the assembled multi-link model for
each data frame such that the difference between the
measured and model-determined marker coordinates

& q&
q q&& =
q

This procedure produces consistent data. However,


the biomechanical model is driven by motion
histories calculated from the inconsistent input data,
which are not the true driven motion histories.
B. Local optimization method
2

13th World Congress in Mechanism and Machine Science, Guanajuato, Mxico, 19-25 June, 2011

since such perturbations typically contain the same


frequencies as those of the movement, each frequency ik
was scaled to be between 0 and 25 rad/s [14].
Once the perturbed movements were defined, the three
procedures described above were applied to these
trajectories, comparing the effectiveness in removing the
mathematical noise.

are globally minimized in a least squares sense, for


all the body segments.
The minimization problem can be stated as an
optimization problem subjected to the kinematic
constraints:

min

s.t.

2
j

j =1

(m (q ) m )

= min

j =1

k =1

jk

jk

(7)

(q ) = 0

2
ij

nb

= min
i=1

(A ( p )r
ni

j=1

i
OMj

IV. Mechanical model


The lower extremity of the human body was modeled in
this study as a multibody system composed of rigid solids
linked by ideal joints, a procedure widely used to analyze
normal and pathological human gait based on inverse
dynamics. Fig. 1 shows a schematic drawing of the
model adopted here. The use of rigid solids to model
limbs is reasonable for large movements that do not
contain strong impacts such as it occurs during normal
walking. In this case, the effects of movements of the soft
tissues around the bones are small.

where n is the total number of markers in the system


and q the coordinates vector of the whole system.
Because of the use of Euler parameters, some
constraints must be added in order to get realistic
parameters:
n
min mj
j=1
s.t.
(q) = 0
1 eij 1

IMD-123

T
i
+ rOi mj ) (Ai ( pi )rOMj
+ rOi mj )

(8)
i [1, nb] j [0,3]

= 1 i [1, nb]

j=0

where nb is the number of segments.


III. Numerical experiments
To provide a basis for the comparison of the three
proposed methods, computer experiments were
performed in order to set a reference solution. A nonperturbed three-dimensional reference movement was
generated by solving the kinematic problem with selected
driving coordinates. Starting from this unperturbed
reference movement, different perturbed movements
were generated by introducing artificial noise into each
three-dimensional marker coordinate. Since Cappozzo et
al. [3] found that skin-fixed markers move in a
continuous rather than random fashion relative to their
underlying anatomical landmarks, a continuous noise
model was chosen,
mik =

mikref

+ noiseik =

mikref

+ Aik cos(ik t + ik )

Aik [0,0.01]m
ik [0,25]H

i [1, n] k [1,3]

Fig. 1. Scheme of the mechanical model adopted

The joints that connect the segments are modeled as


ideal (frictionless) joints, which is not very realistic.
Articular joints present complex contact surfaces as well
as variations of the position of the instantaneous center of
rotation between adjacent segments. However, for large
movements, such as those involved in walking, joints can
be modeled as rotation pairs with a fixed instantaneous
center of rotation, or variable with small errors. This
study used ideal spherical pairs to model the hip, knee
and ankle joints.
The lower extremity model adopted in this study
consists of three rigid bodies representing the thigh, leg
and foot. The system is described by twenty-one
generalized coordinates, seven for each body, describing
the motion of its center of mass and the rotation of the
local system attached to it. The masses, positions of the
centers of mass, and moments of inertia were obtained

(9)

ik [0,2 ]

where mikref is the component k of the position vector of


the marker i estimated by solving the constraint
equations, noiseik is the mathematical noise introduced in
the coordinate k of marker i to simulate the skin
movement and mik the k-component of the position
vector of the perturbed marker, Aik is the amplitude of the
noise, ik the frequency, t the simulated time, and ik the
phase angle. The parameters Aik, ik, and ik were
selected as random numbers, scaled to represent the
motion artifacts anticipated during motion. Since skin
and soft tissue perturbations as large as 2 cm have been
observed experimentally, each amplitude Aik was scaled
to be between 0 and 1 cm (i.e. a 2 cm range). Similarly,
3

13th World Congress in Mechanism and Machine Science, Guanajuato, Mxico, 19-25 June, 2011

IMD-123

from the literature [15] and linearly scaled to fit the


subject studied [16]. All anthropometric data are shown
in Table I.
Data
Length (m)
Proximal CM in Z (m)

Thigh
0.46
0.19

Shank
0.41
0.18

Foot
0.1
0.05

Proximal CM in X (m)

0.117

TABLE 1. Anthropometric data of a subject taken


from [15]
V. Results
In order to solve the problem of kinematic inconsistency,
the three procedures studied here were implemented on
the described mechanical model. Figs. 2-4 show the
temporal evolution of the coordinates of the centers of
mass obtained with the four procedures by imposing the
movement with the analytical equations and adding the
random error as a way of comparing the effectiveness of
each procedure. Figs. 5-8 show the temporal evolutions
of the Euler parameters calculated with the four
procedures in the same way. This effectiveness has been
analyzed in a quantitative way by calculating the RMS
error for each procedure (see Table 2). This RMS has
been normalized by the averaged value of each
component obtained with the analytical equations.

Fig. 3. Temporal evolution of the component y of the center of


gravity of the different segments. Solution with no mathematical
noise. Adaptation model. Local optimization. Global optimization.

Fig. 4. Temporal evolution of the component z of the center of


gravity of the different segments. Solution with no mathematical
noise. Adaptation model. Local optimization. Global optimization

Fig. 2. Temporal evolution of the component x of the center of


gravity of the different segments. Solution with no mathematical
noise. Adaptation model. Local optimization. Global optimization.

13th World Congress in Mechanism and Machine Science, Guanajuato, Mxico, 19-25 June, 2011

Fig. 5. Temporal evolution of the Euler parameter e0 of the different


segments. Solution with no mathematical noise. Adaptation model.
Local optimization. Global optimization

IMD-123

Fig. 7. Temporal evolution of the Euler parameter e2 of the different


segments. Solution with no mathematical noise. Adaptation model.
Local optimization. Global optimization

Fig. 6. Temporal evolution of the Euler parameter e1 of the different


segments. Solution with no mathematical noise. Adaptation model.
Local optimization. Global optimization

Fig. 8. Temporal evolution of the Euler parameter e3 of the different


segments. Solution with no mathematical noise. Adaptation model.
Local optimization. Global optimization

13th World Congress in Mechanism and Machine Science, Guanajuato, Mxico, 19-25 June, 2011

Gen. coordinate
xthigh
ythigh
zthigh
xshank
yshank
zshank
xfoot
yfoot
zfoot

AM
4.39
0.50
40.60
6.85
0.33
32.08
33.61
0.48
33.05

LO
4.14
1.18
9.22
0.63
0.29
4.13
3.72
0.39
1.93

IMD-123

[3] Capozzo A., Catani F. and Leardini A. Skin movement artifacts in


human movement photogrammetry. Proceedings of the XIVth
Congress of the International Society of Biomechanics, pages 238239, 1993.
[4] Cerveri P., Pedotti A. and Ferrigno G. Kinematical models to
reduce the effect of skin artifacts on marker based on human
motion estimation. Journal of Biomechanics, 38:2228-2236, 2005.
[5] Chiari L., Della Croce U., Leardini A. and Capozzo A. Human
movement analysis using stereophotogrammetry. Part 2:
Instrumental errors. Gait and Posture, 21:197-211, 2005.
[6] Leardini A., chiari L., Della Croce U. and Capozzo A. Human
movement analysis using stereophotogrammetry. Part 3: Soft tissue
artifact assessment and compensation. Gait and Posture, 21:212225, 2005.
[7] Silva M.P.T. and Ambrsio J.A.C. Kinematic data consistency in
the inverse dynamic analysis of biomechanical systems. Multibody
Systems Dynamics, 8:219-239, 2002.
[8] Lu T.W. and OConnor J.J. bone position estimation from skin
marker coordinates using global optimization with joint constraints.
Journal of Biomechanics, 32:129-134, 1999.
[9] Capello A., Capozzo A., La Palombara P.F., Luchetti L. and
Leardini A. Multiple anatomical landmark calibration for optimal
bone pose estimation. Human Movement Science, 16:259-274,
1997.
[10] Fuller J., Liu L.J. and Murphy M.C. a comparison of lower
extremity skeletal kinematics measured using skin- and pinmounted markers. Proceedings of the Fourth International
Symposium on 3-D Analysis of Human Movement, International
Society of Biomechanics, Grenoble, France, 1996.
[11] Alonso F.J., Cuadrado J., Lugrs U. and Pintado P.A. A compact
smoothing-differentiation and projection approach for the
kinematic data consistency of biomechanical systems. Multibody
Systems Dynamics, 24:67-80, 2010.
[12] Reinbolt J.A., Schutteb J.F., Fregly B.J., Kohc B., Haftkab R.T.,
Georgec A.D. and Mitchell K.H. Determination of patient-specific
multi-joint kinematicmodels through two-level optimization.
Journal of Biomechanics, 38:621-626, 2005.
[13] Capello A., Francesco P, Palombara, L. and Leardini A.
Optimization and smoothing techniques in movement analysis.
International Journal of Biomedical computation, 41:137-151,
1996.
[14] Chze L., fregly B.J. and Dinnet J. A solidifaction procedure to
facilitate kinematic analyses based on video system data. Journal of
Biomechanics, 28:879-884, 1995.
[15] Ackermann M. Dynamics and energetic of walking with
prostheses. PhD Thesis. University of Stuttgart, 2007.
[16] Ward S.R. and Kingsbury T. Scaling of joint mechanics and muscle
architecture in the human knee. Proceedings of the American
Society of Biomechanics meeting, 2007.

GO
0.57
0.20
8.8
0.75
0.17
3.79
4.14
0.24
2.43

TABLE 2. RMS error in the three procedures


normalized by the averaged value of each component of
the center of gravity for the different segments calculated
with analytical equations without mathematical noise.
In light of the results obtained, global optimization
clearly provides the best results not only in qualitative
way but also in quantitative terms. Local optimization
renders worse results because solving each solid
separately does not take into account what happens in the
rest of the system and therefore some information is lost.
The method of Ambrosio and Silva provides the worst
results alghouth it is the easiest one to implement and it
requires a lower computational effort due to it does not
solve an optimization problem. This suggests that
optimization techniques may be a better approach to
solve the kinematic consistency problem in terms of
accuracy of the solution.
VI. Conclusions
The present work implements a new approach to solve
the kinematic consistency problem using optimization
techniques and Euler parameters. The results obtained in
a theoretical motion show that the method is more
efficient compared with other approaches found in the
literature.
As future work, the new method will be applied to
experimental measurements of human walking. Along
with that, velocity and acceleration analyses will be
carried out in order to study this approach with more
deepness.
Acknowledgements
This work was funded by the Ministerio Espaol de
Educacin y Ciencia as a part of project DPI2009-11 792
and by the Junta de Andaluca as a part of Excellence
Project TEP03115.
References
[1] Siegler S. and Liu W. Three dimensional analysis of human
locomotion, Wiley 1997.
[2] Seireg A. and Arvikar R.J.A mathematical model for evaluation of
forces in lower extremities of the musculo-skeletal system. Journal
of Biomechanics, 6:313326, 1973.
6

13th World Congress in Mechanism and Machine Science, Guanajuato, Mxico, 19-25 June, 2011

IMD-123

Anda mungkin juga menyukai