Volume 2, Issue 2, February 2012 ISSN: 2277 128X

International Journal of Advanced Research in

Computer Science and Software Engineering
Research Paper
Available online at: www.ijarcsse.com
Object Tracking Based on Pattern Matching
V.Purandhar Reddy
Associate Professor, Dept of ECE,
S V College of Engineering,
Tirupati-517501

Abstract In this paper a novel algorithm for object tracking in video pictures, based on edge detection, object extraction and pattern
matching is propsed. With the edge detection, we can detect all objects in images no matter whether they are moving or not. Using
edge detection results of successive frames, we exploit pattern matching in a simple feature space for tracking of the objects.
Consequently, the proposed algorithm can be applied to multiple moving and still objects even in the case of a moving camera. We
describe the algorithm in detail and perform simulation experiments on object tracking which verify the tracking algorithms
efficiency.




I. INTRODUCTION
The moving object tracking in video pictures has attracted
a great deal of interest in computer vision. For object
recognition, navigation systems and surveillance systems,
object tracking is an indispensable first-step.
The conventional approach to object
tracking is based on the difference between the current
image and the background image. However, algorithms
based on the difference image cannot simultaneously
detect still objects. Furthermore, they cannot be applied to
the case of a moving camera. Algorithms including the
camera motion information have been proposed previously,
but, they still contain problems in separating the
information from the background.
In this paper, we propose edge
Detection based Mathod for object tracking in video
pictures. Our algorithm is based on Edge detection, object
extraction and pattern matching. With the edge detection,
we can extract all objects in images. The proposed method
for tracking uses pattern matching between successive
frames. As a consequence, the algorithm can
simultaneously track multiple moving and still objects in
video pictures.
This paper is organized as follows. The
proposed method consisting of stages edge detection,
objects extraction, features extraction & object tracking is
described in detail.

II. PROPOSED CONCEPT FOR MOVING OBJECTS
TRACKING
A. Edge Detection
A problem of fundamental importance in
image analysis is edge detection. Edges characterize object
boundaries and are therefore useful for segmentation,
registration, and identification of objects in scenes. Edge
points can be thought of as pixel locations of abrupt gray-
level change.

In developed algorithm gradient operator method is
used. For digital images these operators also called masks,




represent finite difference approximations of either the
orthogonal gradients f
x,
f
y
or the directional gradient f/r.
let H denote a p p mask and define, for
an arbitrary image U, their inner product at location (m,n)
as the correlation <U,H>
m,n
~
i

I
h(i, j) u(i+m, j+n) =
u(m,n) h ( -m,-n ) here the symbol represents the
convolution.
Let us consider the pair of sobel masks



And


boxed element indicates the location of an object .The
masks H
1
,H
2
measures the gradient of the image U(m,n) in
two orthogonal direction .

Defining the bidirectional gradients

g
1
(m, n) ~ <U,H
1
>
m,n


g
2
(m, n) ~ <U,H
2
>
m,n

the gradient vector magnitude and direction is given by
g (m, n) = ( g
1
(m,n)

2
+ g
2
(m,n)

2
)
1/2

u
g
(m, n)= tan
-1
(g
2
(m,n)

/ g
1
(m,n))
Often the magnitude gradient is calculated as

g (m, n) ~ g
1
(m,n)+g
2
(m,n)
Volume 2, issue 2, February 2012 www.ijarcsse.com
2012, IJARCSSE All Rights Reserved

. This calculation is easier to perform and is preferred
especially when implemented in digital hardware.

The sobel operator computes horizontal and
vertical differences of local sums. This reduces effect of
noise in the data. Noting that this operator have a desirable
property of yielding zeros for uniform regions.

The pixel location (m, n) is declared an edge
location if g(m, n) exceeds some threshold t. The
locations of edge points constitute an edge map (m,n),
which is defined as

Where
I
g
~ { (m,n) ; g (m, n) > t }

The edge map gives necessary data for
tracing the object boundaries in an image. Typically,t
may be selected using the cumulative histogram of g(m, n)
so that 5 to 10% of pixels with largest gradients are
declared as edges.



B. Boundary Extraction By Connectivity Method

Boundaries are linked edges that
characterize the shape of an object. They are useful in
computation in computation of geometry features such as
size or orientation. For extracting the boundaries of an
object connectivity method is used.

Conceptually, boundaries can be found by
tracing the connecting edges. On a rectangular grid a pixel
is said to be four- or eight-connected when it has the same
properties as one of its nearest four or eight neighbors,
respectively as shown in Fig.2(a ,b). There are difficulties
associated with these definitions of connectivity, as shown
in fig.2(c). Under four-connectivity, segments 1, 2, 3, and
4 would be classified as disjoint, although they are
perceived to form a connecting ring. Under eight-
connectivity these segments are connected, but the inside
hole (for example pixel B) is also eight-connected to the
outside (for instance, pixel C). Such problems can be
avoided by considering eight-connectivity for object and
four-connectivity for background. An alternative is to use
triangular or hexagonal grids, where three- or-six-
connectedness can be defined. However, there are other
practical difficulties that arise in working with non
rectangular grids.






Fig .2: connectivity on a rectangular grid .Pixel A and its
(a) 4- connected and (b) 8- connected neighbours;
c) connected paradox; are B and c connected ?

C. Feature Extraction For Objects
In this subsection, we describe the
extracted features of Extracted objects. Figure.3 shows an
example of a object for explanation purposes.
1)Area: By counting the number of pixels included in
object i of the t-th frame, we calculate the area of the
object a
i
(t).

Fig.3: Explanation of the proposed feature extraction from the object
extraction result.
2)Width And Height: We extract the positions of the pixel
P
xmax
(P
xmin
) which has the maximum (minimum) x-
component:
P
xmax
= (X
max,x
,X
max,y
),
P
xmin
= (X
min,x
,X
min,y
),




where X
max,x
, X
max,y
, X
min,x
, and X
min,y
are
the x- and y coordinates of the rightmost and leftmost
boundary of object i, respectively. In addition, we also
extract
P
ymax
= (Y
max,x
, Y
max,y
),
P
ymin
= (Y
min,x
, Y
min,y
).
1 0

1
(m, n)c I
g


thr

(m,n) =

Volume 2, issue 2, February 2012 www.ijarcsse.com
2012, IJARCSSE All Rights Reserved
Then we calculate the width w and the height h of the
objects as follows
w
i
(t) = X
max,x
X
min,x
,
h
i
(t) = Y
max,y
Y
min,y.


3) Position: We define the positions of each object in the
frame as follows
x
i
(t) = (X
max,x
+ X
min,x
)/2
y
i
(t) = (Y
max,y
+ Y
min,y
)/2

4)Color: Using the image data at P
xmax
, P
xmin
, P
ymax
and
P
ymin
, we define the color feature of each object for the R
(Red) component from original color frame
R
i
(t) = [R(P
xmax
) + R(P
xmin
) +R(P
ymax
) + R(P
ymin
)] /4,
as well as by equivalent equations for the G and B
components.

D. Objects Tracking And Distance Measure

The proposed algorithm for object tracking exploits pattern
matching with the features above and makes use of the
minimum distance search in the feature space. We now go
into more details of our algorithm.
Using the edge detection result of the
object i in the t-th frame, we first extract the features of the
object (N+1, i). Here, the notation (N+1, i) stands for the
objects i in the t-th frame. Then we perform the minimum
distance search in the feature space between (N+1, i) and
(N, j) for all objects j in the preceding frame. Finally, the
object (N+1, i) is identified with the object in the
preceding frame which has the minimum distance from
(N+1,i). Repeating this matching procedure for all objects
in the current frame, we can identify all objects one by one
and can keep track of the objects between frames.

Further refinements of the proposed algorithm are in order.
(1) We have not specified the distance measure used for
matching yet. In the simulation experiments we could
confirm that besides the Euclidean distance D
E
the
simpler Manhattan distance D
M
is already sufficient
for object tracking purposes.
(2) In order to treat all object features with equal weights,
it is necessary to normalize the features. One possible
way is dividing them by their maximum values.
Dividing by 2
n
, where the integer n is determined for
each feature so that approximately equal weights
results, is another possibility. The second possibility
has the advantage that the division can be realized by
a shifting operation in a hardware realization. Figure 4
shows a block diagram of proposed method & Figure
5 shows a Detailed description of the proposed object
tracking algorithm.

Fig. 4: Block diagram of proposed object tracking method.

[Objects Tracking Algorithm]
1) Convert the color image to gray scale image
2) Perform the edge detection by sobel edge
detection.
3) Dilate the image by boundary connectivity.
4) Extract all objects by labeling method
5) Feature Extraction
a) Extract the Features(Area, width, height &
color features) of object to track in N
th

Frame(ie previous frame)
b) Extract the Features (Area, width, height &
color features) of object to track in N+1
th

Frame( ie previous frame
6) Pattern Matching in the Feature
Space
a) Calculation of distances
Search for the minimum
Distance among the distances.
b) Apply Feature match of N
th
Frame object
with minimum distance object of N+1
th

Frame object if not matched perform the
feature match next minimum distance object
and so on .
c) After matching remove the data of N
th

Frame and store the data of N+1
th
Frame.
d) Increase the value of N by N+1 Repeat the
steps 1 to 6
.
Fig. 6: Detailed description of the proposed object tracking algorithm.
III. SIMULATION

The proposed algorithm is tested using Matlab 7.1.
For experimental verification two different video
Volume 2, issue 2, February 2012 www.ijarcsse.com
2012, IJARCSSE All Rights Reserved
sequences were taken from moving camera. Then frames
were extracted from the video sequences. Since all the
processing is done on gray scale images, 24 bit color
image frame is initially converted into gray scale frame of
8 bit size. By giving frames one by one to the Matlab
program of the proposed algorithm, the tracked object
segmentation is extracted. The dimension of the processing
image is 320240.

Frame
number
Object
tracked
result
Result of
the tracked
object

FRAME 1


----------


------------

FRAME 2

Pattern
matched


FRAME 3
Pattern
Matched


FRAME 4

Pattern
matched


FRAME 5
Pattern
matched

Fig. 7: The tracked object results from successive frames.

TABLE-I: EXTRACTED FEATURES FOR SAMPLE 1
F.
No
A W H P R G B

1

5

3.2

2.2

4.3

6.1

4.6

5.1

0.03

2

4.75

2.9

2.2

4.3

6.1

4.4

4.8

0.03

3

4.2

2.7

2.1

4.2

6.1

4.3

4.7

0.03

4

3.95

2.7

2

4.1

5.9

4.7

5.1

0.03

5

3.5

2.6

2.1

4

5.9

4.2

4.4

0.03

6

3.21

2.5

2

4

5.9

4.2

4.5

0.03

7

2.8

2.3

2

4

5.9

4.3

4.7

0.03

8

2.6

2

2.1

4.1

5.9

4

4.3

0.03
IV. CONCLUSION
We have proposed an object tracking algorithm for
video pictures, based on Edge detection and pattern
matching of the Extracted objects between frames in a
simple feature space. Simulation results for frame
sequences with moving objects verify the suitability of the
algorithm for reliable moving object tracking. We also
have confirmed that the algorithm works very well for
more complicated video pictures including rotating objects
and occlusion of objects.
In order to extract color features of Extracted
objects, we used four boundary pixels color features from
the original image. Thus, correct color features of an object
that has gradation or texture is not extracted. Nevertheless,
the mean value turns out to sufficiently represent the
objects color features for the tracking purpose.


REFERENCES


[1] A new appearance model based on object sub-region for tracking
Shu-
Peng Wang;Hong-BingJi; Wavelet Analysis and Pattern
Recognition, 2007. ICWAPR '07.

[2] A moving object tracking method based on sequential detection
scheme Qiu Xuena; Liu Shirong; Liu Fei; Zhu
Weitao;DuFangfang;Control Conference (CCC), 2010 29th Chinese.

[3] A stream field based partially observable moving object tracking
algorithm Kuo-ShihTseng;Control, Automation, Robotics and
Vision, 2008. ICARCV 2008. 10th International Conference.

[4] W. G. Kropatsch and H. Bischof, Digital Image Analysis,
Springer,
2001.

[5] G.L. Foresti, A real-time system for video surveillance of
unattended
outdoor environments, IEEE Trans. Circuits and Systems for Vid.
Tech., Vol. 8, No. 6, pp. 697704, 1998.

[6] C. Stauffer and W.E.L. Grimson, Learning patterns of activity
using
real-time tracking, IEEE Trans. PAMIL, Vol. 22, No. 8, pp. 747
757,
2000.
[7] H. Kimura and T. Shibata, Simple-architecture motion-detection
analog V-chip based on quasi-two-dimensional processing, Ext.
Abs.
of the 2002 Int. Conf. on Solid State Devices and Materials
(SSDM2002), pp. 240 241, 2002.

[8] S. W. Seol et al., An automatic detection and tracking system of
moving objects using double differential based motion estimation,
Proc.
of Int. Tech. Conf. Circ./Syst., Comput. and Comms. (ITC-
CSCC2003),
pp. 260 263, 2003.

[9 ] An embodiment of stereo vision system for mobile robot for real-
time
measuring distance and object tracking Ik-Hwan Kim; Do-Eun Kim;
You-Sung Cha; Kwang-hee Lee; Tae-Yong Kuc;
Control, Automation and Systems, 2007. ICCAS '07. International
Conference
[10] Image Segmentation and Pattern Matching Based FPGA/ASIC
Implementation Architecture of Real-Time Object Tracking
K. Yamaoka, T. Morimoto, H. Adachi, T. Koide, and H. J.
Mattausch
Research Center for Nanodevices and Systems, Hiroshima
University.


.