Available online at www.sciencedirect.

com

Gait & Posture 28 (2008) 588595

www.elsevier.com/locate/gaitpost

Joint kinematics estimate using wearable inertial and magnetic

sensing modules
Pietro Picerno a,*, Andrea Cereatti a,b, Aurelio Cappozzo a
a

Department of Human Movement and Sport Sciences Istituto Universitario di Scienze Motorie, Piazza Lauro de Bosis 6, 00194 Roma, Italy
b
Centro di Cura e Riabilitazione Santa Maria Bambina, Oristano, Italy
Received 15 October 2007; received in revised form 10 March 2008; accepted 7 April 2008

Abstract
Background and aims: In many applications, it is essential that the evaluation of a given motor task is not affected by the restrictions of the
laboratory environment. To accomplish this requirement, miniature triaxial inertial and magnetic sensors can be used. This paper describes an
anatomical calibration technique for wearable inertial and magnetic sensing modules based on the direct measure of the direction of
anatomical axes using palpable anatomical landmarks. An anatomical frame definition for the estimate of joint angular kinematics of the
lower limb is also proposed.
Methods: The performance of the methodology was evaluated in an upright posture and a walking trial of a single able-bodied subject. The
repeatability was assessed with six examiners performing the anatomical calibration, while its consistency was evaluated by comparing the
results with those obtained using stereophotogrammetry.
Results: Results relative to the up-right posture trial revealed an intra- and inter-examiner variability which is minimal in correspondence to
the flex-extension angles (0.22.98) and maximal to the internalexternal rotation (1.67.38). For the level walking, the root mean squared
error between the kinematics estimated with the two measurement techniques varied from 2.5% to 4.8% of the range of motion for the flexextension, whereas it ranged from 13.1% to 41.8% in correspondence of the internalexternal rotation.
Conclusion: The proposed methodology allowed for the estimate of lower limb joint angular kinematics in a repeatable and consistent
manner, enabling inertial and magnetic sensing based systems to be used especially for outdoor human movement analysis applications.
# 2008 Elsevier B.V. All rights reserved.
Keywords: Movement analysis; Anatomical calibration; Anatomical frame definition; Joint angular kinematics; Wearable devices; Inertial and magnetic
sensing

1. Introduction
The description of joint kinematics during the execution
of a physical exercise may be accomplished by tracking the
trajectory of active or passive point markers located on the
skin of the subject using optoelectronic stereophotogrammetry (SP). The reconstructed coordinates of these markers
in an arbitrarily defined global frame (GF) allow for the
determination of local frames (technical frame: TF)
associated with each bone of interest and, therefore, the
description of the relevant instantaneous pose (position and
* Corresponding author. Tel.: +39 06 36733506; fax: +39 06 36733517.
E-mail address: pietro.picerno@iusm.it (P. Picerno).
0966-6362/$ see front matter # 2008 Elsevier B.V. All rights reserved.
doi:10.1016/j.gaitpost.2008.04.003

orientation). The use of this instrumentation is constrained

by the fact that the measurement volume is limited and that
the relevant equipment is financially demanding. A possible
alternative to this approach is the use of motion analysis
systems based on electromagnetic [1,2] or ultrasound [3]
technologies. But again, the ultrasound or electromagnetic
fields generated by these systems are limited in volume.
More recently, the availability of miniature solid-state
inertial and magnetic sensors has opened up a new
perspective [4,5]. A three-dimensional (3D) linear accelerometer, a 3D angular rate sensor and a 3D magnetometer
have been assembled to produce a sensor module, referred to
as magnetic field angular rate and gravity sensor (MARG
[6]). The use of specific sensor fusion algorithms can

P. Picerno et al. / Gait & Posture 28 (2008) 588595

provide an estimate of the orientation of a TF embedded in

the module housing relative to a global Earth-fixed GF
defined by the direction of gravity and the local magnetic
North vector [7,8]. This can be accomplished with no spatial
limitations and at a relatively low cost. The estimate of
the global position of the TF is not yet reliable since the
double integration of the linear acceleration, which yields
displacement data, can be accurately calculated only for
types of motion where the velocity of the MARG sensor
returns cyclically to zero [9]. In addition, a further limitation
may be encountered. The presence of metallic objects in the
vicinity of a MARG sensor can change the direction of the
local magnetic North vector and therefore the orientation of
the GF thereby introducing an artefact. The latter
disturbances that may affect the sensors involved may be
compensated by the above-mentioned sensor fusion algorithms for limited intervals of time [7]. These problems
may, in turn, limit the maximal duration of the movement
tracking.
For an effective and repeatable representation of angular
kinematics, an anatomical frame (AF) must be associated
with each bone involved in the analysis [10,11].
Since this frame is normally different from the relevant
TF, a transformation matrix from the latter frame, provided
by the motion-tracking equipment, to the AF must be
determined (movement-morphology registration). Relevant
information is collected through a procedure referred to as
anatomical calibration.
In regards to the above stated purpose, superficial
anatomical landmarks (ALs) are identified by manual
palpation and their location relative to the TF are determined
using an ad hoc experiment [12]. The location of internal
ALs may be determined either as a function of the location
of superficial ALs or, when applicable, as the mean centre of
rotation of a selected movement between two adjacent bony
segments [13]. A functional approach may also be used for
the identification of anatomical axes (AAs). A bone may be
made to rotate, relative to its adjacent counterpart, in an
anatomical plane. The relevant mean axis of rotation is taken
as the AA orthogonal to that plane [14]. The utilization of the
functional approach to determine both ALs or AAs may be
open to dispute since, for its reliability, the mechanical
behaviour of the joint involved must be very close to that of a
spherical or a cylindrical hinge, respectively, which is rarely
the case. In addition, the joint involved must be capable of
sufficiently wide angular excursions [15].
Different geometric rules have been proposed that define
AFs using AL positions and AA orientations [12,14,16,17]
determined as illustrated above. Although it has been
demonstrated that joint kinematics numerical representation
depends highly on AFs definition [18] and, consequently, the
movement analysis community considers it a priority, a final
consensus in this respect has not yet been reached.
Conversely, given the AFs of two adjacent bones, it is
now common practice to estimate joint kinematics using the
so-called Cardan convention [19].

589

Differently from motion analysis system based on stereophotogrammetry, electromagnetic or ultrasound technologies, using MARG sensors, the position of ALs cannot be
determined, while the orientation of axes is provided. The
axes of the AF can be assumed coinciding with the axes of
the TF by manually aligning the sensor module with the
anatomical axes. This approach is evidently unreliable
since it only approximates anatomical planes and is not
repeatable. A second solution, suggested by several authors
[20,21], is based on a functional approach: one of the axes of
the AF can be assumed as coinciding with the direction of
the 3D angular velocity vector measured by the MARG
sensor attached to the body segment while the segment is
rotated about a joints functional axis. A second axis of the
AF can be defined using the direction of gravity measured by
the MARG sensor module during resting posture. This
approach has the advantage of being quick to perform (so it
is ideal for virtual reality and entertainment applications),
but undergoes some limitations: movements planes and
postures are subjective and, thus, the determination of the
AFs may be not repeatable. Moreover, joint impairments
may produce axes that are not consistently related to bone
anatomy.
The purpose of this study is to present a novel anatomical
calibration procedure to be used in association with a
MARG based motion-tracking system. This procedure is
based on the identification of superficial ALs. The proposed
methodology can be used to compute 3D joint angular
kinematics by means of MARG sensors in a reliable manner.
The repeatability of the method and its consistency with
joint kinematics obtained using SP are assessed with
reference to the lower limb during posture and level walking.
2. Materials and methods
The anatomical calibration is carried out while the subject
assumes a suitable stationary posture with the body segments of
interest equipped with a movement tracking MARG sensor and
using a calibration device. The latter is equipped with two mobile
pointers that can be made to point two-selected ALs. The body of
the device carries a MARG sensor, an active axis of which is
aligned with the line joining the two pointers and, thus, two ALs. In
this way the orientation of this line relative to the GF can be
measured. For each bony segment, the orientation of a minimum of
two non-parallel lines must be determined in order to construct the
orthogonal axes of the AF through a given geometric rule. Calibration devices of different geometry may be required in order to fit
bones with different shapes (Fig. 1).
During the anatomical calibration procedure, the MARG sensor
module attached to the body segment provides the orientation
matrix (gRt(0)) of the TF while the MARG sensor module placed
on the calibration device measures the orientation of at least two
unit vectors (guk; k = 1, 2) of lines passing through couples of ALs,
all relative to the GF. The latter vectors are, thereafter, represented
in the TF through the rigid transformation
t

uk g RTt 0  g uk

(1)

590

P. Picerno et al. / Gait & Posture 28 (2008) 588595

Fig. 1. Calibration devices adopted to identify the lines connecting ALs (a and b). Calibration of the directions identified by GT and LE (c), and of the direction
identified by ME and LE with respect to the thigh TFMARG (d).

and used to calculate the orientation matrix of the AF in the TF

(tRa(0)) through a geometric rule.
During the execution of a physical exercise, in each i-th sampled
instant of time, the motion-tracking MARG sensor provides the
matrix gRt(i), i = 1, . . ., N and the global orientation of the AF,
necessary to estimate joint kinematics. It is calculated as
g

Ra i g Rt it Ra 0;

i 1; . . . ; N

(2)

The MARG sensor based anatomical calibration procedure

described above can only rely on external palpable ALs. Thus,
the most commonly adopted AF definitions for femur, tibia and foot
cannot be used [12,14,16,17]. The ALs and the geometric rules
used in this study that define the AFs for pelvis and lower limb bony
segments are illustrated in Fig. 2. It should be emphasized that other
definitions are difficult to realize.
During the experiments, both the MARG sensor outputs and the
stereophotogrammetric data (9 VICON Mx cameras, Oxford
Metrics, UK), were sampled at 120 samples per second. To minimize problems related to ferro-magnetic disturbances, experiments
were conducted in a controlled magnetic field environment. A
single able-bodied adult subject was involved in the experiment.
Four MARG sensors (MTx, Xsens Motion Technologies,

Enschede, The Netherlands) were firmly attached to the volunteers

sacrum, the latero-distal thigh, the medial facet of the tibia and,
laterally, to the tarsal bones. Each module was equipped with a
three retro-reflective point marker cluster. The two markers on the
rod were 10 cm apart from each other and approximately 7 cm from
the marker placed on the case (Fig. 1c and d).
The AL locations, listed in Fig. 2, were identified by an expert
through manual palpation and marked with a felt pen. Six examiners performed the MARG sensors anatomical calibration procedure using the above-described calibration devices. One of the
examiners performed the anatomical calibration procedure six
times. Each calibration session took approximately 5 min.
Thereafter, retro-reflective markers were placed on the ALs
using the relevant markings.
Since the AFs definitions adopted in this study differed from
those commonly adopted in gait analysis, the kinematics estimate
was compared with that obtained using a selected commonly
adopted lower limb AFs definition [12]. For this purpose, extra
markers were located on the left posterior iliac spine, the tibial
tuberosity and the second metatarsal head. The position of these
markers in the TFs constructed using the relevant point marker
clusters were determined through SP. After this calibration was
carried out, the markers that indicated the ALs were removed. The

P. Picerno et al. / Gait & Posture 28 (2008) 588595

591

Fig. 2. Right lower limb anatomical frame definitions. uk represents the oriented axes identified by pointing two ALs by means of the calibration devices. The
origins of the AFs are arbitrarily placed.

hip joint centre in the pelvis point marker cluster was also determined using the functional approach described in [13].
An experimental trial was carried out while the volunteer
assumed an upright posture. A second experimental trial was
performed during level walking at a self-selected speed. During
both trials the global orientation of the MARG sensor and the
global position of the point marker clusters were tracked simultaneously.
The joint angular kinematics of hip, knee and ankle, during
upright posture and during level walking, were estimated using
both SP and MARG sensor data. In both cases the Cardan angular
convention and the AF definitions reported in Fig. 2 were used.
Data from SP were also used to estimate the joint angular
kinematics according to commonly adopted AF definitions [12].
2.1. Data analysis
The intra- and inter-examiner repeatability of the MARG sensor
anatomical calibration procedure was evaluated in terms of root
mean square deviation (RMSD) from the mean of the joint angles
estimated during the up-right posture and generated by the six
relevant calibrations.

The consistency between the two experimental methods was

assessed by comparing, for each joint, the angles obtained from the
data provided by the MARG sensors, using the anatomical calibrations performed by the six examiners, and the angle provided by the
SP system. To this purpose, the absolute value of the differences
between the former six angles and the latter angle was calculated
and the relevant descriptive statistics (mean and standard deviation)
determined. This was done using the data collected during the
upright posture.
With reference to the level walking trial, consistency was
evaluated both in terms of offset and waveform dissimilarity among
the joint angles time histories. The offset was calculated as the
mean absolute difference (MAE) between the arithmetic mean of
each of the curves obtained with the MARG sensors and that
obtained with SP. To analyze the waveform dissimilarity, curves
obtained with the MARG sensors and SP were aligned with respect
to their relative mean value. From the aligned curves, the root mean
square error (RMSE), averaging over one gait cycle, and the RMSE
expressed as a percentage of the nominal joint rotational amplitude
(RMSE%) were derived.
The comparison between the AF definitions adopted in this
study and that proposed by [12] was carried out by calculating the

592

P. Picerno et al. / Gait & Posture 28 (2008) 588595

correlation coefficient (R) and difference in the range of motion

(DRoM) between the joint angular kinematics estimated using the
two different AF definitions.

Table 1
Intra- and inter-examiners repeatability (RMSD) of joint angles estimated
with MARG sensors during an upright posture
Joint

3. Results
Precision and consistency in estimating joint angles
during the up-right posture trial using the MARG sensors are
reported in Tables 1 and 2. On average, the intra-examiner
repeatability was higher than inter-examiner repeatability. In
particular, the internalexternal rotation (In/Ex rot) variability was the highest across all joints. The knee In/Ex rot
exhibited the highest intra- and inter-examiners RMSD, 4.98
and 7.38, respectively.
For all joints, the flexion-extension angle (Fl/Ex)
presented the smallest AE, with the mean values ranging
from 1.38 to 1.88 and standard deviation values below 0.98.
The largest AE was observed in correspondence with the In/
Ex rot and, in particular, the highest mean value of AE, equal
to 8.38, was found for the ankle, whereas the largest S.D.
value of AE, equal to 6.18, was observed in correspondence
with the hip.
For all joints, the MAE affecting the angular kinematics
estimated by the MARG sensors during one gait cycle was
minimal for the Fl/Ex angle and maximal for the In/Ex rot.
For Fl/Ex, it ranged from 1.28 at the ankle to 38 at the hip.
For the Ab/Ad, it ranged from 3.68 at the hip to 5.58 at the
ankle and finally for the In/Ex rot, it varied from 4.58 at the
hip to 21.78 at the ankle. The RMSE values varied from 0.88
to 3.68 and were minimal for the Fl/Ex and maximal for the

Hip
Knee
Ankle

Fl/Ex (8)

Ab/Ad (8)

In/Ex rot (8)

Intra

Inter

Intra

Inter

Intra

Inter

2.9
0.2
0.4

1.8
2
1.5

1.2
1
1

2.4
1.1
1.5

3.5
4.9
1.6

6.6
7.3
1.6

Table 2
Mean (standard deviation) of the differences in absolute value (AE) between
joint angles estimated with SP and the MARG sensors during an upright
posture
Joint

Fl/Ex (8)

Ab/Ad (8)

In/Ex rot (8)

Hip
Knee
Ankle

1.8 (0.7)
1.9 (0.7)
1.3 (0.9)

3 (2.2)
4.6 (1.1)
5.7 (1.5)

6.7 (6.1)
6.3 (3.9)
8.3 (1.6)

In/Ex rot, across all joints and angles. The Fl/Ex curves
estimated with the MARG sensors were very similar to those
estimated with SP (Fig. 3) and the RMSE were always lower
than 4.8% of the nominal RoM (Table 3). The largest
RMSE% value was the knee In/ex rot and was equal to
41.8% of the nominal RoM.
The joint kinematics obtained using the AF definitions
proposed in this study were highly correlated with those
estimated using the AF definitions suggested by [12].
The correlation coefficient for the Fl/Ex was equal to 1 for
all joints whereas the DRoM was less than 0.58. The
lowest R was the knee In/Ex rot, and it was equal to 0.942

Fig. 3. Ensemble plots of hip, knee and ankle joint angle waveforms. The solid black lines represent the joint kinematics obtained with SP and the dashed grey
lines represent the joint kinematics obtained with the MARG sensors as generated by the anatomical calibrations performed by the six examiners. The curves are
temporally normalized from initial right foot contact (0%) until subsequent foot contact (100%).

P. Picerno et al. / Gait & Posture 28 (2008) 588595

Table 3
Comparison of joint angular kinematics estimate relative to one gait cycle
obtained with the MARG sensors and SP
Joint

Angle

RMSE (8)

RMSE %

Hip

Fl/Ex
Ab/Ad
In/Ex rot

MAE (8)
3
3.6
4.5

0.8
1.5
1.8

2.5
13.3
13.1

Knee

Fl/Ex
Ab/Ad
In/Ex rot

2.4
4.8
9.4

1.9
2.8
3.6

3
21.1
41.8

Ankle

Fl/Ex
Ab/Ad
In/Ex rot

1.2
5.5
21.7

1.2
2.2
3.5

4.8
11.7
32.5

The offset was evaluated in terms of mean absolute error (MAE). The
waveform distortion was evaluated in terms of root mean square error
(RMSE) between joint angular kinematics curves aligned with respect to
their relative mean values, estimated with SP and the MARG sensors.
RMSE % represented the RMSE expressed as a percentage of the RoM
estimated with SP.

whereas the largest DRoM, equal to 28, was the knee Ab/
Ad (Fig. 4).

4. Discussion and conclusion

The proposed methodology for the anatomical calibration
using MARG sensors led to a consistent and repeatable
estimation of the lower limb AFs. The joint angles computed
during an upright posture were used to evaluate the
differences in the orientation of the AFs determined using
the MARG and SP and the repeatability of the calibration
procedure performed with the MARG sensors. The estimate

593

of the Flex/Ex angles was the most reliable for all joints
both in terms of repeatability (RMSD < 2.98) and in
terms of consistency with the SP data (AE < 1.98 (0.78)).
The estimates of the Ab/Ad angles were repeatable
(RMSD < 2.48) but were associated to differences with
respect to the SP data up to 5.78 (1.58). The less reliable
angles estimate was for the In/Ex rot which exhibited both
the lowest repeatability (RMSD < 7.38) and consistency.
To exclude errors related to the uncertainties associated
with ALs identifications, their positions were identified by
an expert and then marked with a felt pen. Thanks to this,
two objectives were accomplished: all operators performed
the anatomical calibration exercise by pointing at the same
previously marked ALs; the AFs, determined using the
MARG sensors and SP, were defined using the same points,
thus allowing comparison between the results. Unfortunately, SP reconstructs the position of the geometrical centre
of the marker placed on the AL, which is different from the
palpable position of the AL on the skin. Hence, the direction
of the line connecting two ALs measured with a MARG
sensors (hosted in its specific calibration device) and with SP
is unavoidably different. This difference increases when the
distance between two ALs decreases (i.e. for epicondyles
and malleoli), resulting in an AF determined differently
from one anothers measurement system even if identical
ALs and axis definitions are used. This circumstance may
explain the greater errors found in correspondence with the
Ab/Ad and In/Ex rot.
It is important to note that the cluster of markers was
mounted on the MARG sensor case, and therefore, during the
walking trial, TFs obtained with MARG sensors and SP were
affected by the same soft tissue artefact and were related by a

Fig. 4. Ensemble plots of hip, knee and ankle joint angle waveforms. The solid lines represent the joint kinematics obtained using a commonly adopted AF
definition [12], while the dashed lines represent the joint kinematics obtained from the AF definition suggested in this study. The movement data have been
recorded using SP.

594

P. Picerno et al. / Gait & Posture 28 (2008) 588595

time-invariant rigid transformation. The performance of the

methodology was not evaluated with respect to the actual
bone movement, which was unknown, but with respect to the
movement obtained by SP, considered as the gold standard.
The joint angular kinematics obtained with the MARG
sensors was comparable to that obtained using SP. The Fl/Ex
angles estimate showed the highest accuracy (RMSE < 28,
MAE < 38). The highest errors were found in correspondence
with the In/Ex rot angles. In general, the kinematic patterns
were quite well reproduced (RMSE < 3.68) whereas the
offset from the curves obtained with SP were high and in
particular for the ankle joint (MAE = 21.78).
The AF definitions chosen in the present work differ from
those commonly adopted in gait analysis. In fact, the AF
definitions were conditioned by the fact that only superficial
ALs can be pointed using the calibration device, which
provides axis orientation rather than position of points.
However, the estimated joint kinematics was still consistent
(0.942 < R < 1, 0.58 < DRoM < 28) with the conventional
description of the joint movement.
In general, gait analysis requires gathering information
on the spatio-temporal parameters and the joint kinematics
and kinetics. The aim of this study was limited to the
improvement of the quality of joint angular kinematics
estimate with MARG sensors. Different methods for the
determination of spatio-temporal parameters from accelerations and angular velocities have been described [22]. To
the authors knowledge, only one study faced the problem of
joint kinetics estimation [23], analyzing solely ankle
moment and power. Future research should be aimed at
the development of a methodology, based on the use of
MARG sensors, which allows for ambulatory analysis of the
lower limb kinetics.
A limitation in the use of this type of instrumentation is
that the presence of ferromagnetic disturbances can affect
the orientation of the MARG sensors. In fact, in a nonhomogeneous magnetic environment each MARG sensor
computes its orientation with respect to a different GF. This
results in an erroneous estimate of the relative movement
between MARG sensors. It is, hence, crucial to operate in
controlled conditions during both the calibration procedure
and the acquisition trial.
In conclusion, the proposed methodology, based on the
direct identification of ALs, enables MARG sensor based
systems to be used as a valid alternative to SP for three
dimensional joint angular kinematics estimate, especially in
those applications where the evaluation of the motor task
should be carried out in a natural environment and for a
prolonged interval.

Acknowledgements
The authors would like to thank Andrea Mele for his
technical support. This study was funded by the Italian
Ministry of Health and the ISPESL.

Conflict of interest
None.

References
[1] Kobayashi K, Gransberg L, Knutsson E, Nolen P. A new system for
three-dimensional gait recording using electromagnetic tracking. Gait
Posture 1997;6:6375.
[2] Mills P, Morrison S, Lloyda DG, Barrett RS. Repeatability of 3D gait
kinematics obtained from an electromagnetic tracking system during
treadmill locomotion. J Biomech 2007;40:150411.
[3] Kiss RM, Kocsis L, Knoll Z. Joint kinematics and spatial-temporal
parameters of gait measured by an ultrasound-based system. Med Eng
Phys 2004;26:61120.
[4] Kemp B, Janssen AJ, van der Kamp B. Body position can be monitored
in 3D using miniature accelerometers and earth-magnetic field sensors. Electroencephalogr Clin Neurophysiol 1998;109:4848.
[5] Zhu R, Zhou Z. A real-time articulated human motion tracking using
tri-axis inertial/magnetic sensors package. IEEE Trans Biomed Eng
2004;12:295302.
[6] Bachmann EB. Inertial and magnetic tracking of limb segment
orientation for inserting humans into synthetic environments. Ph.D.
Thesis. Naval Postgraduate School, Monterey, CA; 2000.
[7] Roetenberg D, Luinge HJ, Baten CT, Vltink PH. Compensation of
magnetic disturbances improves inertial and magnetic sensing of
human body segment orientation. IEEE Trans Neural Syst Rehabil
Eng 2005;13:395405.
[8] Sabatini AM. Quaternion-based strap-down integration method for
applications of inertial sensing to gait analysis. Med Biol Eng Comput
2005;43:94101.
[9] Veltink PH, Slycke P, Hemssems J, Buschman R, Bultstra G, Hermens
H. Three dimensional inertial sensing of foot movements for automatic
tuning of a two-channel implantable drop-foot stimulator. Med Eng
Phys 2003;25:218.
[10] Cappozzo A. Gait analysis methodology. Hum Movement Sci
1984;3:2750.
[11] Lafortune MA, Cavanagh PR, Sommer HJ, Kalenak A. Three-dimensional kinematics of the human knee during walking. J Biomech
1992;25:34757.
[12] Cappozzo A, Catani F, Della Croce U, Leardini A. Position and
orientation of bones during movement: anatomical frame definition
and determination. Clin Biomech 1995;10:1718.
[13] Camomilla V, Cereatti A, Vannozzi G, Cappozzo A. An optimized
protocol for hip joint centre determination using the functional
method. J Biomech 2006;39:1096106.
[14] Frigo C, Rabuffetti M, Kerrigan DC, Deming LC, Pedotti A. Functionally oriented and clinically feasible quantitative gait analysis
method. Med Biol Eng Comput 1998;36:17985.
[15] Woltring HJ, Huiskes R, de Lange A, Veldpaus FE. Finite centroid and
helical axis estimation from noisy landmark measurements in the
study of human joint kinematics. J Biomech 1985;18:37989.
[16] Davis RBI, Ounpuu S, Tyburski D, Gage JR. A gait data collection and
reduction technique. Hum Movement Sci 1991;10:57587.
[17] Wu G, Siegler S, Allard P, Kirtley C, Leardini A, Rosenbaum D, et al. ISB
recommendation on definitions of joint coordinate system of various
joints for the reporting of human joint motion. Part I. Ankle, hip, and
spine. International Society of Biomechanics. J Biomech 2002;35:5438.
[18] Della Croce U, Camomilla V, Leardini A, Cappozzo A. Femoral
anatomical frame: assessment of various definitions. Med Eng Phys
2003;25:42531.
[19] Grood ES, Suntay WJ. A joint coordinate system for the clinical
description of three-dimensional motions: application to the knee. J
Biomech Eng 1983;105:13644.

P. Picerno et al. / Gait & Posture 28 (2008) 588595

[20] Luinge HJ, Veltink PH, Baten CT. Ambulatory measurement of arm
orientation. J Biomech 2007;40:7885.
[21] ODonovan KJ, Kamnik R, OKeeffe DT, Lyons GM. An inertial and
magnetic sensor based technique for joint angle measurement. J
Biomech 2007;40:260411.

595

[22] Aminian K, Najafi B, Bula C, Leyvraz PF, Robert P. Spatio-temporal

parameters of gait measured by an ambulatory system using miniature
gyroscopes. J Biomech 2002;35:68999.
[23] Schepers HM, Koopman HF, Veltink PH. Ambulatory assessment of
ankle and foot dynamics. IEEE Trans Biomed Eng 2007;54:895902.