Documente Academic
Documente Profesional
Documente Cultură
8, AUGUST 2005
AbstractA new method of measuring joint angle using a combi- the progress of new technologies, powerful microcontrollers,
nation of accelerometers and gyroscopes is presented. The method miniature sensors, high capacity memories and small bat-
proposes a minimal sensor configuration with one sensor module teries have been developed; allowing to realize low-power and
mounted on each segment. The model is based on estimating the ac-
celeration of the joint center of rotation by placing a pair of virtual portable recording systems carried by the subject for long-term
sensors on the adjacent segments at the center of rotation. In the ambulatory measurements. Therefore there has been much
proposed technique, joint angles are found without the need for in- activity in trying to find alternative solutions for capturing gait
tegration, so absolute angles can be obtained which are free from information over a larger distance and outside a laboratory
any source of drift. The model considers anatomical aspects and
is personalized for each subject prior to each measurement. The environment. During the last decade, body-mounted sensors
method was validated by measuring knee flexion-extension angles consisting of accelerometers and/or rate gyroscopes have been
of eight subjects, walking at three different speeds, and comparing used to obtain kinematic values such as shank and thigh in-
the results with a reference motion measurement system. The re- clination angles, and knee joint angle [6], [7]. These data can
sults are very close to those of the reference system presenting very
be derived by integration of angular acceleration or angular
small errors (rms = 1 3, mean = 0 2, SD = 1 1 deg ) and
excellent correlation coefficients (0.997). The algorithm is able to velocity. However, data obtained from integration can be dis-
provide joint angles in real-time, and ready for use in gait anal- torted by offsets or any drifts [8], [9]. The two major sources
ysis. Technically, the system is portable, easily mountable, and can of drift are: 1) Electronic bias error [10], and 2) Deviation and
be used for long term monitoring without hindrance to natural turning out from the sensing axis. To eliminate any drift during
activities.
integration, Morris [11] identified the beginning and the end of
Index TermsAccelerometer and gyroscope, ambulatory the walking cycles, and made the signal at the beginning and
system, biomechanics, gait analysis, joint angle. the end of the cycle equal. Tong et al. [2] applied a low-cut
high-pass filter on the shank and thigh inclination angle signals.
I. INTRODUCTION Time-frequency analysis (Wavelet Transform) was also applied
to lower limb angular velocity in order to remove the drift [3].
(4)
Fig. 1. A physical sensor module at point P , and a virtual sensor module on (5)
point C on a 2-D rigid body. Each sensor module consists of 2-D accelerometers
and a gyroscope.
where and are the first and second derivatives of angle of
with respect to a fixed inertial frame.
II. METHODS The two latter parameters can be measured by the gyroscope
A. Model Description placed at point . The virtual gyroscope reading at point will
Considering the two-dimensional (2-D) model of a segment also give the same signal [2]. So considering and equal to
motion (Fig. 1), a sensor module including a biaxial accelerom- and , respectively, (4) can be rearranged to yield the virtual ac-
eter and a gyroscope was placed at point on the segment. celerometer readings with respect to known physical accelerom-
The first step was to calculate the expected signals of a virtual eters and gyroscope readings
sensor module placed at an arbitrary point on the bone seg-
ment with an arbitrary orientation with respect to the physical (6)
sensor module. The vector indicates the distance, and the
angles and represent the orientations of the physical and The next step was to calculate joint angle between two
virtual sensors with respect to . segments using two modules of sensors mounted on each seg-
To estimate the virtual signals, a description of the outputs of ment (Fig. 2). Thus, the two physical sensor modules on each
the sensors was required. A single axis accelerometer measures segment were shifted to the joint center, or more precisely the
the difference of acceleration and gravity along its sen- center of rotation point, such that each virtual sensor aligns with
sitive axis given by the unit vector . The measured electrical its corresponding segment orientation. Since one point should
signal ( ) could thus be expressed as physically have a unique acceleration, the two virtual sensors
meeting at the center of rotation should give equal accelerations.
(1) However, the correction for coordinate frames rotation by angle
should be considered
Similarly, by considering a 2-D accelerometer with sensitive
(7)
axes along and , the two measured signals were given by
where and are virtual ac-
celerometers readings at point on segments 1 and 2, respec-
(2) tively. These vectors can be rewritten in polar representation
(10)
B. Test Protocol
Eight healthy subjects, who had given informed consent, par-
ticipated in this experiment, five men and three women, aged (11)
between 44 and 70 yr . The volunteers per-
formed three 30 s flat treadmill walking trials at speeds 2, 3, and
4 km/h, wearing their basket shoes. One of the subjects was also The angle was obtained from the dot product formula
requested to perform a freely arbitrary flexion and extension of
knee, such as sitting, standing, and swinging. (12)
To calculate the knee angle (flexion-extension) using the
proposed model, two sensor modules, each containing two
accelerometers and one gyroscope, were used. The sensors where , and
(dimension: ) were mounted on .
the left shank and thigh segments using a strap [Fig. 3(a)]. The The angle was obtained by calculating the difference be-
sensing axes were adjusted in the anterior-posterior plane so tween inclination angles of the kinematic sensor and the
that the flexion-extension angle could be measured. All signals
vector
were sampled at 200-Hz using the Physilog [BioAGM, CH]
ambulatory system carried on the waist [6].
Before the walking trials, three small markers were pasted (13)
over the left lateral malleolus , the lateral epicondyle ( ),
and the junction of the first and second proximal lateral third of where and are averages of the 2D accelerometer read-
the thigh [Fig. 3(b)]. Then each subject was asked to stand ings during the standstill trial.
in two positions at extended [Fig. 3(b)] and flexed [Fig. 3(c)] left Similarly, the values of and were obtained by changing
knee positions, while the shank was kept stationary. At each po- all indices 1 to 2 in (12) and (13).
sition the subject stayed for a few seconds ( 5 s) at standstill, For comparison, a Zebris CMS-HS (Zebris, D) ultra-
while the system was recording kinematic parameters, a lateral sound-based motion measurement system was used as the
view photograph was taken as well. The cameras image plane reference system [14]. Three Zebris markers were used that
was adjusted to be in parallel with sagittal plane to avoid per- were attached over the same adhesive markers ( , , and
spective errors. The position and orientation of the camera was C). Spatial marker positions (x, y, z) were recorded with an
kept constant so that the two photos would have the same field absolute accuracy better than 1.0 mm [15], [16] and used for
of view. The known length of the metal frame (70 cm height) calculation of knee flexion-extension angle. Synchronization
was used to calibrate the photos from pixels to metric units (cm). between the reference system and the Physilog systems was
Theses images were used to estimate the coordinates of markers performed automatically.
DEJNABADI et al.: ACCURATE MEASUREMENT OF UNIAXIAL JOINT ANGLES 1481
Fig. 3. (a) Attachment of the kinematic sensors on both thighs and shanks using straps. The kinematic data are recorded by the Physilog system (dimensions
13 cm 2 7 cm 2 3 cm, weight: 300 gr) placed in the waist bag. (b) Sagittal view of a subject with extended left knee and (c) with flexed left knee, representing
the sensor configurations. The white circles indicate the position of markers (M , C and M ), and the white squares indicate the position of kinematic sensors (P
and P ).
C. Data Analysis
Matlab was used for all signal processing. A third-order
SavitzkyGolay filter [17] was applied to smooth the ac-
celerometers and gyroscopes signals. Anthropometry data
obtained by photography were also fed to the model to estimate
the expected virtual sensors readings shifted to the knee center
of rotation.
For comparison with the reference system, the error signal
was defined as the difference between the angle obtained by the
proposed method and the reference system.
Statistical analyses comprised rms, mean, and standard devia-
tion of difference error, as well as correlation coefficient calcula-
tions between the joint angles obtained by the proposed method
and the reference systems data.
III. RESULTS
Figs. 47 show the steps of calculating knee angle of Fig. 4. Physical accelerometers and gyroscopes (raw data) readings during
subject no. 8 during a flat walking at 3 km/h. Fig. 4 shows walking at 3 km/h. The site on thigh consists of two accelerometers (S and
the physical accelerometers and gyroscopes (raw data) read- S ) and a gyroscope (! ). Similarly, there are two accelerometers (S and
S ) and a gyroscope (! ) on shank module.
ings placed on thigh and shank. Fig. 5 shows the virtual
accelerometers readings placed at the knee center of rotation
on the adjacent segments. Fig. 6 shows the polar representa- measured by the reference system, and difference error between
tion (modulus, argument) of the same virtual accelerometers, the two results.
as well as the difference between the two moduli signals Fig. 8 shows also the final step of calculating knee angle and
. Since both and express its comparison with the reference system during a freely arbi-
the accelerations of the same point , the value of the difference trary flexion and extension of knee (
is very small . Fig. 7(a) shows the final step , ).
yielding the knee angle calculated from the difference between The whole results are summarized in Table I, which outlines
the two argument signals . the mean and standard deviation of difference errors, together
Fig. 7(b) shows the knee angle calculated from position data as with the calculated correlation coefficients of eight subjects. The
1482 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 52, NO. 8, AUGUST 2005
Fig. 8. Absolute knee angle during a freely arbitrary flexion and extension
of knee. (a) Calculated from the new accelerometers and gyroscopes settings.
(b) Calculated from position data as measured by the reference system (solid
line), and difference error between the two results (dashed line).
TABLE I
COMPARISON BETWEEN KNEE ANGLE MEASUREMENTS OBTAINED BY BODY-MOUNTED SENSORS AND ZEBRIS MARKERS FOR 8 SUBJECTS AT
3 SPEEDS. THE ERROR REPRESENTS THE rms, MEAN AND SD OF THE DIFFERENCE SIGNAL BETWEEN ZEBRIS AND OUR MEASURING DEVICE.
r REPRESENTS THE CORRELATION COEFFICIENT BETWEEN THE TWO MEASURING SYSTEMS
Since the angle computation needs simple multiplications and but the information is not enough to correct or compensate for
derivative operations, this system is able to provide real-time the errors. This information is obtained by checking the error
knee angle for any type of activity. So the clinicians are able to difference between the moduli signals of the two virtual ac-
view joint angle motion as the subject performs the prescribed celerometers (Fig. 6).
activity, or generally it can be used in many other applications Technically, the system is portable, easily mountable, and it
that require real-time feedback. can also be used for long-term monitoring without hindrance to
Unlike electrogoniometers, the proposed system provides the natural gait [6]. In comparison with video-based systems,
also anterior-posterior rotations and linear accelerations of this system can be an alternative solution for capturing kine-
thigh and shank independently, which can further be used matic information over a nonlimited distance and outside a lab-
for a better estimation of lower limbs kinematics. Based on oratory environment.
its MEMS technology, gyroscopes and accelerometers also The proposed method was found very promising in providing
offer a cheaper and more practical solution to the cumbersome actual knee flexion-extension angle during daily activities. Al-
electrogoniometer link over the knee as well [18], [19]. though the model proposed estimation of uniaxial joint rota-
The proposed model considers anatomical aspects of each tion (i.e., flexion-extension), the principle could be extended to
subject prior to each measurement that leads to higher accuracy multiaxial joint rotation by employing 3-D-accelerometers and
in the results. In this way, a better orientation of bone segments 3-D-gyroscopes on each site.
(femur and tibia) can be estimated from the sensors placed on Future prospects should consist in estimating lower limb tra-
the skin as it was described in Fig. 2 and (10)(12). jectory in daily environment and extending the model to 3-D.
However, there are some limitations in the model due to as-
suming the joint center of rotation as a fixed position point, and ACKNOWLEDGMENT
segments as rigid bodies. Although the joint center of rotation The authors are grateful to J. Gramiger and P. Morel in the
changes slightly depending on the knee angle [20], this effect in- design of the ambulatory system; and C. Voracek for her assis-
duces very minimal changes in the knee joint angle. The model tance in collecting the data.
allows a uniaxial (flexion-extension) estimation of joint angle;
however it does not constrain the joint motion to take place REFERENCES
in sagittal plane: any flexion-extension even out of the sagittal
[1] E. Medri, D. Tepavac, B. Needham, and D. Popovic, Comprehensive
plane can be estimated. gait analysis in spinal cord injured patients with functional electrical
Skin motion artefact, a common source of error to all body stimulation, in Proc. Ann. Conf. IEEE Eng. Med. Biol. Soc., 1994, pp.
mounted devices, affects the measurement accuracy. The thigh 357858.
[2] K. Tong and M. H. Granat, A practical gait analysis system using gy-
sensor is more susceptible to skin and soft tissue artifact where roscopes, Med. Eng. Phys., vol. 21, pp. 8794, 1999.
the majority of the femur is concealed by a substantial amount [3] K. Aminian, C. Trevisan, B. Najafi, H. Dejnabadi, C. Frigo, E. Pavan, A.
of soft tissue. The effect of skin artefact was minimized by using Telonio, F. Cerati, E. C. Marinoni, P. Robert, and P.-F. Leyvraz, Evalu-
adequate elastic band to fix the sensors, and applying low-pass ation of an ambulatory system for gait analysis in hip osteoarthritis and
after total hip replacement, Gait & Posture, vol. 20, pp. 102107, 2004.
filtering on the raw signal. [4] R. B. Davis, S. Ounpuu, D. Tyburski, and J. R. Gage, A gait analysis
From a practical standpoint, misalignment of the sensors or data collection and reduction tequnique, Hum. Mov. Sci., vol. 10, no. 5,
sensor deviation during movement reduces the systems accu- pp. 7587, 1991.
[5] L. M. Schutte, U. Narayanan, J. L. Stout, P. Selber, J. R. Gage, and M.
racy. Fortunately, there is a redundant information in the system H. Schwartz, An index for quantifying deviations from normal gait,
which cab be used to check the overall effects of error sources, Gait Posture, vol. 11, pp. 2531, 2000.
1484 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 52, NO. 8, AUGUST 2005
[6] K. Aminian, B. Najafi, C. Bla, P.-F. Leyvraz, and P. Robert, Spatio- Hooman Dejnabadi received the B.Sc. and M.Sc.
temporal parameters of gait measured by an ambulatory system using degrees in electrical engineering from Sharif Univer-
miniature gyroscopes, J. Biomech., vol. 35, pp. 689699, 2002. sity of Technology, Tehran, Iran, in 1993 and 1996 re-
[7] R. E. Mayagoitia, A. V. Nene, and P. H. Veltink, Accelerometer and rate spectively. He is currently working towards the Ph.D.
gyroscope measurement of kinematics: an inexpensive alternate to op- degree in biomedical engineering at the Laboratory of
tical motion analysis systems, J. Biomech., vol. 35, pp. 537542, 2002. Movement Analysis and Measurement, Ecole Poly-
[8] K. Aminian and B. Najafi, Capturing human motion using body-fixed technique Fidirale de Lausanne (EPFL).
sensors: outdoor measurement and clinical applications, J. Visual He was with Fara Sanat Shomal Co. and
Comput. Animat. Comput. Animation and Virtual Worlds, vol. 15, pp. RadRavesh Co. from 1993 to 2001 as a Research
7994, 2004. Engineer on NDT systems. His research interests
include biomedical signal processing, ambulatory
[9] A. T. M. Willemsen, J. A. Van Alste, and H. B. K. Boom, Real-time
systems, and movement analysis.
gait assessment utilizing a new way of accelerometry, J. Biomech., vol.
23, no. 8, pp. 859863, 1990.
[10] D. H. Titterton and J. L. Weston, Strapdown Inertial Navigation Tech-
nology. Stevenage, U.K.: Peregrinus, 1997, pp. 178179.
[11] J. R. W. Morris, Accelerometerya technique for the measurement Brigitte M. Jolles received the MSc degree in mi-
of human body movements, J. Biomech., vol. 6, no. 7, pp. 2936, crotechnology engineering from the Swiss Federal
1973. Institute of Technology (EPFL) in 1990, the M.D.
[12] A. Heyn, R. E. Mayagoitia, A. V. Nene, and P. H. Veltink, The kine- degree from University of Lausanne (UNIL) in 1995,
matics of the swing phase obtained from accelerometer and gyroscope and the specialist degree in orthopedic surgery. She
measurements, in Proc. 18th Ann. Conf. IEEE Eng. Med. Biol. Soc., received the Diploma in Clinical Epidemiology from
vol. 2, 1996, pp. 463464. the University of Toronto (UofT), Toronto, ON,
[13] K. Halvorsen, M. Lesser, and A. Lundberg, A new method for esti- Canada, in 2002.
mating the axis of rotation and the center of rotation, J. Biomech., vol. She is an Orthopaedic Surgeon, an Engineer with
32, pp. 12211227, 1999. the Swiss Federal Institute of Technology (EPFL),
[14] R. M. Kiss, L. Kocsis, and Z. Knoll, Joint kinematics and spatial-tem- and Assistant Professor at the UNIL. In 2001-2002,
poral parameters of gait measured by an ultrasound-based system, Med. she worked as a Clinical and Research Fellow in the Department of Orthopedic
Eng. Phys., vol. 26, pp. 611620, 2004. Surgery at UofT. She became Assistant Professor at UNIL in February 2005.
[15] H. M. Overhoff, D. Lazovic, M. Liebing, and C. Macher, Total She is part of the surgeon staff at the Orthopaedic Hospital of the University of
knee arthroplasty: coordinate system definition and planning based on Lausanne and teaches at the pregraduate and postgraduate levels at UNIL and
3-D ultrasound image volumes, in Proc. 5th Annu. Conf. Int. Soc. other institutions. Her research interests include joint arthroplasty and ambula-
Computer-Aided Surgery, vol. 1230, Int. Congress Series, 2001, pp. tory systems for movement analysis.
292299.
[16] Zebris Medizintechnick GmbH, Measuring System for 3D-Motion
Analysis CMS-HS, Tech. Data and Operating Instructions Text Release
8/99, 1999.
Kamiar Aminian (M89) received the electrical
[17] A. Savitzky and M. J. E. Golay, Analyt. Chem., vol. 36, pp. 16271639,
engineering degree in 1982 and the Ph.D. degree
1964. in biomedical engineering in 1989 from Ecole
[18] C. M. Myles, P. J. Rowe, C. R. C. Walker, and R. W. Nutton, Knee Polytechnique Fidirale de Lausanne (EPFL).
joint functional range of movement prior to and following total knee He worked as a Research Associate in the
arthroplasty measured using flexible electrogoniometry, Gait Posture, Metrology Laboratory, EPFL, and as an Assistant
vol. 16, pp. 4654, 2002. Professor at Sharif University of Technology,
[19] A. Shiratsu and H. J. C. G. Coury, Reliability and accuracy of different Tehran, Iran. In January 2002, he joined the School
sensors of a flexible electrogoniometer, Clin. Biomech., vol. 18, pp. of Engineering of EPFL where he is the Head of
682684, 2003. the Laboratory of Movement Analysis and Measure-
[20] J. D. Moorehead, S. C. Montgomery, and D. M. Harvey, Instant center ment. He teaches in the areas of sensors and medical
of rotation estimation using Reuleaux technique and a lateral extrapola- instrumentation. His research interests include transducers, movement analysis,
tion technique, J. Biomech., vol. 36, pp. 13011307, 2003. ambulatory systems, and biomedical signal processing