2-D Comparative Gait Kinematics Using a Single Video Camera and EMG Signal Analysis

taupeselectionΜηχανική

14 Νοε 2013 (πριν από 3 χρόνια και 7 μήνες)

241 εμφανίσεις

2
-
D Comparative Gait Kinematics Using a Single Video Camera
and EMG Signal Analysis


Sujeet Blessing C.
#1
, Chaitanya Srinivas L.V.
*2



#
M. Tech. Student, Biomedical Engineering Division, School of Biosciences and Technology,

VIT University, Vellore, Tamil

Nadu, India


*
Assistant Professor, Biomedical Engineering Division, School of Biosciences and Technology,


VIT University, Vellore, Tamil Nadu, India


_______________________________________________________________________________
_________



ABSTRACT


Wal
king has been most comprehensively studied of all human movements. Last two decades have
witnessed the increase in technical advancements to a whole new level in the ability to collect and
process an enormous amount of data. As the advancements increases,
so does the cost. The
initiative of this project is to come up with a low cost gait analysis method which would use simple
techniques yet be efficient as the standard ones. Electromyography of nine muscles from the leg is
acquired using biokit physiograph
during one gait cycle. Kinematic analysis has been done using
Marker and Marker
-
less Video
-
Graphic techniques. Image, Video and Signal Processing are carried
out using MATLAB R2009b. To avoid frame
-
by
-
frame analysis each time, both the techniques are
made
automated in a way that when the video of the subject’s gait is fed, program would run for
each frames and give out the final output. For kinematics, a single video camera is used to capture
the subjects’ gait in 2
-
D view. The outcomes of both the marker b
ased and marker
-
less video
-
graphic analysis are compared. Stride analysis data are collected for all the subjects using Paper
-
Ink method.


Keywords:

Gait kinematics, Marker based video
-
graphic technique, Marker
-
less video
-
graphic
technique, EMG analysis.


INTRODUCTION:


Human locomotion, or gait, may be described as a translator progression of the body as a
whole, produced by coordinated, rotator movements of body segments. Walking is efficiently done
with the coordinated work of joint mobility and muscle
activity that is selective in timing and
intensity. It is a complex activity since, it is dependent on a series of interactions between two multi
-
segmented lower limbs and total body mass.

There are many systems to measure the parameters
related to gait. H
owever, the main reason why the gait analysis is not quite popular among the
commercial purpose
s

is the cost of the software and hardware that one requires to do the analysis
effectively.

The main criterion is to compare marker and marker
-
less video
-
graphi
c techniques,
which have been developed having cost in mind and works as efficient as the standard techniques
that are available in market.

Marker technique is widely being used in every place regardless of the
tedious work one has to go through in doing
the analysis. Marker
-
less technique is getting more
attention these days because of its easier way to carry on the gait analysis. So, comparison of
these two techniques is essential to validate marker
-
less technique is as standard as the marker
technique.
EMG from nine leg muscles have been acquired to know the activity period, onset and
offset of activity and power of each muscle on a particular joint during locomotion. By comparing
these EMG parameters with the kinematics data, it is possible to give surg
ical recommendation for
the patients before undergoing a surgery.


DETERMINANTS OF POSITIONS:


Electromyography obtained from muscles associated with the joint movements during
normal gait is an active determinant of that particular joint motion. The body
moves due to the
energy that has been generated by means of concentric contraction of muscle groups and this
generated energy is removed by eccentric contractions with each steps. Gravity or movement of the
opposite side of the body are some forces which a
re considered as passive determinants of the
joint motion which constitutes kinetics and it is not being dealt in this paper.


METHADOLOGY

A.

EMG ACQUISITION:


EMG is acquired for six normal subjects from nine lower extremity muscles during walking
on a unif
orm surface. BIOKIT Physiograph is used to acquire EMG. Surface electrodes are
preferred over needle electrodes due to its non
-
invasiveness. The electrodes used are bi
-
polar
electrodes which can efficiently acquire both the positive and negative signals ge
nerated in the
muscles during a gait cycle. To place the electrodes effectively on the belly of the muscle,
“NOROXAN


EMG and Sensor Systems’ Clinical SEMG Electrode Sites” chart has been used.
The subjects have been asked to shave the hair in and around
the region of the electrode sites to
get a proper communication between electrode and the skin. The signals thus obtained are less
noisy compared to those signals obtained from non
-
shaved leg. Once the electrodes are placed in
the respective positions, EMG

signal is recorded for one gait cycle.

The anterior lower extremity muscles from which EMG is acquired are Rectus Femoris (RF),
Vastus Lateralis (VL), Vastus Medialis (VM), Soleus (S) and Tibialis Anterior (TA). The posterior
lower extremity muscles from
which EMG is acquired are Biceps Femoris (BF), Semi tendinosus
(ST), Medial Gastrocnemius (ST) and Lateral Gastrocnemius (LG). All these above mentioned
muscles have profound activity in determining the joint positions of both hip and knee.


B.

EMG PROCESSIN
G:




The acquired EMG is processed using MATLAB 7.6 2008a. Linear envelope, Integrated
EMG and Root Mean Square are some parameters found out upon processing the signal. Any
EMG signal should be normalized before processing to avoid errors, because, EMG

signals vary
from subject to subject and from time to time. So, Isometric Maximum Voluntary Contraction (IMVC)
is carried out for each muscle right before the acquisition of EMG signal during walking. EMG is
acquired when the subject is asked to contract
the specific muscle while pulling of pushing against
an immovable object provided the muscle should not let the associated joints move. The peak value
of the IMVC gives the scalar factor. Normalization is done by multiplying the signal with this scalar
fac
tor.




LINEAR ENVELOPE
:



It gives the muscles’ activation and de
-
activation point during one gait cycle. Raw signal is
loaded in MATLAB, base line is shifted to zero, multiplied with the scalar factor corresponding to the
IMVC of the subject, full wave

rectification is done and finally the signal is passed through a low
-
pass Butterworth filter with a cut
-
off frequency of 5 Hz.





Biceps Femoris



Vastus medialis


Vastus Lateralis






Semi Tendinosus


Rectus Femoris




Medial Gastrocnemius






Lateral Gastrocnemius


Soleus



Tibialis Anterior


INTEGRATED EMG:


It gives the total muscle activity of a muscle during one gait cycle. This follows the same
procedure as linear envelope in processing. The

linear envelope of the signal is passed through an
integrator.




Average IEMG of Six Subjects



Average RMS of Six Subjects




ROOT MEAN SQUARE:


EMG signal has both positive and negative criteria. So, in order to assess both, RMS of the
signal is im
perative. RMS gives the power of signal which includes both the positive and negative
sides. It is done by squaring each data points, summing together, divided by the total number of
observations. Square root of the above result gives the power of the EMG
signal.

All the parameters obtained above are averaged for six subjects.


C.

STRIDE ANALYSIS:


Time and distance are two basic variables of motion. The stride analysis variables are most
commonly used to describe a gait pattern. It gives essential quantitat
ive information about a
person’s gait. Each variable could be affected by factors like age, sex, height, size and shape of
bony components, distribution of mass in body segments, joint mobility, muscle strength, type of
clothing and footwear, habit, and ps
ychological status. It is carried out on the same six subjects
from whom the EMG has been acquired. The method used is paper
-
ink, where the subject walks on
a paper following a straight line.

The variables are:

Stride length, Step length, Cadence, Stride w
idth, Velocity and Foot
progression angle.

The variables obtained for each subjects are averaged.

Subjects

Step length
(in cm)

Stride
length (in
cm)

Cadence
(Steps/min)

Stride
width (in
cm)

Velocity (in
m/sec)

Foot
Progression
angle (in
deg)

1

65.5

132

74

10.5

0.8

5.25

2

55.8

114.5

74

10.5

0.68

4.7

3

68.7

138.4

81

19.7

0.92

8.1

4

50.8

102.2

76

6.9

0.64

5.2

5

63.2

127

98

21

1.03

11.3

6

59.8

119.8

102

18.2

1.01

10.2

Average

60.6

123.3

84

14.4

0.84

7.4

Mean Stride Analysis for Six Subjects


D.

KINEMATIC A
NALYSIS:


MARKER BASED VIDEO
-
GRAPHIC TECHNIQUE

It describes the motion of objects without consideration of the causes leading to the motion.
Helen Hayes marker set is used. The markers are affixed to the anatomical positions. A camera
which captures at the

rate of twenty five frames/second is used. The camera is placed at a distance
of 9 feet from the straight line, the subject has been asked to walk. Proper arrangements have been
made to make sure the camera is perpendicular to the straight line. Active ma
rker set is used. LEDs
are preferred over radium, because, former is more illuminating than the later and also during
image processing LEDs are found to be more effective. To make the light not scatter too much,
channeled LEDs are preferred as in a torch l
ight. So, during illumination the markers are seen as
round coloured dots. Since, the markers used are active in nature; a perfect dark environment is
required to do gait analysis. The subjects are asked to walk in a straight line with lights on, until the
y
feel comfortable in doing so in the dark. Once the gait analysis of each subjects are recorded, one
gait cycle (from one heel strike to next heel strike) is extracted from the video and processed.

IMAGE AND VIDEO PROCESSING:

The captured video is loaded
on MATLAB 7.6 2008a and frame
-
by
-
frame analysis is done to
measure the angles of hip and knee flexion
-
extension. Program has been written to each frame
starting from the first.


Let R (a,b,t) be the RGB video.

Where, t = (t
1
,t
2
,t
3
,....,t
N3
), n3 = total nu
mber of frames, a = (a
1
,a
2
,a
3
,....,a
N1
), n1=total number of
columns in the frame t
N3
, b = (b
1
,b
2
,b
3
,.....,b
N2
), n2 = total number of rows in the frame t
N3
.

Each RGB frames in the RGB video is converted into greyscale images with an intensity lying
betwee
n 0 and 255, where 0 being black and 255 being white. Let R
G

(a,b,t) be the greyscale video.

This converted greyscale image is further transformed into binary image with only two possibilities
for each pixel,

either 1 or 0. So, the markers which are affixe
d on the anatomical locations turn into
1’s with a 0’s background. Let R
BW

(a,b,t) be the binary video.

These markers of 1’s are considered as blobs and they are detected and labelled in an
order of fashion from first column to the last column. The order o
f the blobs labelled, could change
unpredictably during the different phases of gait. So, in order to accurately pin point each marker,
spatially they are thresholded with each other and labelled as hip marker, sacrum marker, knee
marker and ankle marker.
After labelling each marker, the centroid for each marker is measured. To
find out the centroid for each marker, edge detection is done, using these border pixels’
coordinates, centroid is obtained using the

below

formula


x
C
, y
C

= ((x
1
+x
2
+x
3
+…..x
N
)/
N
, (y
1
+y
2
+y
3
+…..y
N
)/
N
).


Where, (x
C
, y
C
) = coordinate of a centroid; (x
1
, y
1
), (x
2
, y
2
),…,( x
N
, y
N
) = coordinates of the border of
the blob.


With the centroids of each marker, respective lines are drawn connecting these markers.
For
measuring hip angle, line
is drawn from the centroid of the sacrum marker to the centroid of the
anterior superior iliac spline marker and another line connecting centroid of ASIS and femoral
epicondyle (knee marker). Likewise, for measuring knee angle, line is drawn connecting ASI
S and
femoral epicondyle and another line connecting femoral epicondyle and mallelous. Hough’s
transform is used to detect the lines which are drawn, previously. When the lines are detected,
Hough’s transform gives the angle between the two lines. As expl
ained above the hip and knee
angle for each frame are measured and plotted.


The angle of hip and knee for each frame is individually plotted and averaged for six subjects
.







Hip angle



Knee angle



a


one frame of an original video; b


grey ima
ge; c, d


binary image; e


blob detection; f


for hip angle
estimation; g


for knee angle estimation; h


detected lines by Hough’s transform for hip angle; i


detected
lines by Hough’s transform for knee angle


MARKER
-
LESS VIDEO
-
GRAPHIC ANALYSIS:

For

the last few years, there has been a tremendous advancement in kinematic analysis.
Marker
-
less video
-
graphic analysis has a lot of pros compared to marker based techniques. One
being the best advantage is that the subject does not have to wear any markers

and also the
subject does not have to undress for the analysis. So, the subject can walk comfortably, there is no
need of any subject preparations and the analysis can be done in any environment. This paper
deals with a modeless analysis wherein there is
no need of any predefined model which helps in
analysing a subject’s gait. Camera is placed perpendicular to the straight line where the subject
walks. Same camera which has been used for marker based video
-
graphic technique is used at a
distance of 9 feet

from the straight line.

IMAGE AND VIDEO PROCESSING:

To process this video, the captured video has to be converted into silhouette. To get
silhouette, the walking subject has a same coloured background which could be eliminated. During
video processing th
e background colour is turned black while the rest of the colours into white and
the silhouette of the subject alone is extracted from the background. The distance between ankle to
knee and knee to hip are manually measured for each subjects and these dist
ances are correlated
with the pixels of the image of the corresponding subjects. During stance phase, that is 60% of gait
cycle, length of the leg remains the same for all the frames, but during swing phase the length of leg
tends to get shortened to avoid

dragging the foot on the floor and the knee tends to go a little higher
from mid
-
swing to terminal swing until the heel is placed than during stance phase. So, the distance
of foot raised during the swing phase is calculated and this distance is deduced i
n the marker
levels.

Let R(a,b,t) be the video which is in RGB format. Where, t = (t
1
,t
2
,t
3
,....,t
N3
), N3 = total number of
frames, a = (a
1
,a
2
,a
3
,....,a
N1
), N1=total number of columns in the frame tn3, b = (b1,b2,b3,.....,b
N2
),
N2 = total number of rows
in the frame t
N3
. Let S(a,b,t) be the video of silhouette. Where, the
background is zeros and the subject appearing is ones.

First the leg of the subject is extracted, eliminating the torso. This extracted image starts from the
hip to the base of the foot
.

Let S
ST

(a
ST
,b
ST
,t) be the extracted image of only leg during stance
phase.

First row would be bst1 = b
1
+b
HIP
, and the last row would be b
STN

= b
N2
-

BASE OF THE FOOT
.

Similarly, a
ST1

= a
1

+ Lowest column which has 1’s.


a
STN

= a
1



Highest column w
hich has 1’s.

Hip marker is Mean of (S
ST
(1’s(b
ST1
))),

Ankle marker is Mean of (S
ST
(1’s(b
STN
-
0.125))),

Knee marker is Mean of (S
ST
(1’s(b
STN
-
0.5))).

Let S
SW

(a
SW
,b
SW
,t) be the extracted image of only leg during swing phase.

First row would be b
SW1

= b
1
+b
HIP
,

and the last row would be b
SWN

= b
STN

+b
LG

Let b
LG

= Distance from the floor to the foot’s base. When b
LG

increases, it means swing phase has
started.

Similarly, a
SW1

= a
1

+ Lowest column which has 1’s.


a
SWN

= a
1



Highest column which has 1’s.

From
initial swing to mid swing same algorithm as stance phase is used.

During mid swing,

Hip marker is Mean of (S
SW
(1’s(b
SW1
))),

Ankle marker is Mean of (S
SW
(1’s(b
STN
-
0.125
-
b
LG
))),

Knee marker is Mean of (S
SW
(1’s(b
STN
-
0.5
-
b
LG
))).

During the end of the termina
l swing, when blg is 0, again algorithm for stance phase is used.
After
attaining the position of the markers, lines are drawn connecting the respective markers. Line
detection and angle measurement are carried out using Hough’s transform, the same way as
marker based video
-
graphic technique. The angles of both hip and knee for each frame are plotted.

a


Silhouette of a original frame; b


image extracted from; d


negative image; e


correlating the manual the hip;

c


extracting only the subject from t
he background; measurements with the pixel values; f


shin; g


upper leg;

h


drawing lines connecting the markers; i


detected lines using Hough’s transform

The angle of hip and knee for each frame is individually plotted and averaged for six subjects.




Hip Angle





Knee Angle




DISCUSSION
:

Marker based technique as well as Marker
-
less video
-
graphic techniques have found to give
consistent angle data compared to the standard data. Marker technique has a lot of disadvantages,
starting from,
preparation of the subjects for the analysis, alignment of markers on the exact
anatomical locations, making sure the markers don’t move when the subject walks, inconvenience
brought to the subject because of the markers which could affect his normal gait,

motion blur in less
efficient cameras, uncomfortable feeling for the subjects to remove their clothes off, choosing the
right environment to do the analysis, to conduct the gait analysis in a dark environment when active
markers are used, which makes the
patient not be able to walk in a straight line. On the other hand,
marker
-
less video
-
graphic technique is easier to do. There is no need of any subject preparation,
can walk with his clothes on, no need of any specific environment. With a similar coloured
background, and enough space, marker
-
less technique can be carried out without any difficulties of
complications. Knee angle attained from marker technique found to have a smooth curve compared
to the one that of marker
-
less technique, though, both the gra
phs are up to the mark of the standard
knee angle. Hip angle is almost same for both the marker and marker
-
less video
-
graphic
techniques as well. In marker based system, positioning LEDs play a crucial role in deciding the
angle. Whereas, in marker
-
less te
chnique, the subject is asked to wear a tight fitting pants, so,
errors would be minimal.

CONCLUSION
:

Stride analysis have been carried out on normal subjects. EMG is acquired and processed
for nine leg muscles. An automated cost effective system of Marker

based video
-
graphic technique
and silhouette based marker
-
less video
-
graphic technique have been developed and the results are
compared and found to be consistent with the normal angle data from the literature. As future work,
more than a single video cam
era can be used, integrated and coordinated. So, kinematics could be
done in 3
-
D view. Abduction
-
Adduction and Internal
-
External rotation could be measured with
different anatomical planes. Kinetics could also be integrated along with kinematics and EMG
an
alysis.

REFERENCES
:

[
1] Richard Baker, “Gait analysis methods in rehabilitation”, Journal of NeuroEngineering and Rehabilitation,
2006, 3:4.

[2] Mary M. Rodgers, “Dynamic biomechanics of the normal foot and ankle during walking and running”,
Physical Ther
apy, 1988, 1822
-
30.

[3] Michela Goffredo, Imed Bouchrika, John N. Carter and Mark S. Nixon, “Performance analysis for gait in
camera networks”, Association of Computing Machinery, 2008, 73
-
80.

[4] Y.P. Ivanenko, R.E. Poppele and F. Lacquaniti, “Five basic
muscle activation patterns account for muscle
activity during human locomotion”, American Journal of Physiology, 2004, 267
-
282.

[5] M.B.I. Reaz, M.S. Hussain and F. Mohd
-
Yasin, “Techniques of EMG signal analysis: Detection,
processing, classification and a
pplications”, Biological Procedures, 2006, 8(1): 11
-
35.

[6] Noraxon EMG and Sensor System, “Clinical SEMG Electrode Sites.” www.noraxon.com.

[7] Helen Hayes Marker System, www.helenhayeshospital.org.





____________________________________________________
___________________________


sujeet_blessing2000@yahoo.com

chaitanya.pro@gmail.com