Sei sulla pagina 1di 7

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/254819945

CAMERA OBJECT TRACKING FAST ALGORITHM

Article

CITATION READS

1 44

3 authors, including:

Nikola Teslic
RT-RK Computer Based Systems
121 PUBLICATIONS   662 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

hysteresis phenomenon View project

Human-Device Interface Research View project

All content following this page was uploaded by Nikola Teslic on 05 June 2014.

The user has requested enhancement of the downloaded file.


CAMERA OBJECT TRACKING FAST ALGORITHM
1
Nikola Teslic, 2Vladimir Radenkovic, 3Snezana Crnogorac
1
Novi Sad Institute of Information Technologies MicronasNIIT, Fruskogorska 11a, 21000
Novi Sad, Serbia and Montenegro, Nikola.Teslic@MicronasNIT.com
2
Novi Sad Institute of Information Technologies MicronasNIIT, Fruskogorska 11a, 21000
Novi Sad, Serbia and Montenegro, Vladimir.Radenkovic@MicronasNIT.com
3
Novi Sad Institute of Information Technologies MicronasNIIT, Fruskogorska 11a, 21000
Novi Sad, Serbia and Montenegro, Snezana.Crnogorac@MicronasNIT.com

ABSTRACT the other hand, the face of the person being tracked
must be turn toward the camera during the process.
This paper addresses camera real-time object tracking The other factor that influences the effectiveness of
problems. Here we present one camera object tracking
application is the fact that the color depends on the
solution based on block-matching technique. In our solution
object is marked in the first frame, and afterwards its
general level of light brightness of the area in which
movement is tracked using the block-matching technique in the object is moving.
limited area of the picture. The size of limited area depends Object could also be tracked based on its shape
on predicted object movement between two frames. We are [4]. Human body is not a rigid object, so the body
using block-matching technique in reduced number of pixels, shape during movement changes from frame to frame,
searching for the best matched square blocks in two and that also makes the tracking more difficult.
successive frames. In the first frame, square block is Commonly used object tracking technique is
centered to the object that we are tracking. Square block tracking based on the object motion between two
center difference in two neighboring frames determines
frames. The tracking based on the motion between two
object movement between these two frames. Our algorithm
does not depend on position between object and camera and
frames is just the calculation of differences between
can be used in wide range of light brightness. Object pixel characteristics of two successive frames [3]. If
tracking is possible even when other objects are moving near human face is being tracked, it is difficult to separate
object of interest. Algorithm is implemented on PC platform the movements of the face from the movements of the
with minimal additional hardware resources. Stepper motor rest of the body. On the other hand, in the observed
is used for camera movement. area there are many objects that are not of interest, and
that are difficult to separate from the object of interest.
1. INTRODUCTION With this technique there is one additional problem,
during the tracking, the camera moves and that creates
Object or human tracking is a complex problem in the movement of the still area that must be
field of image processing. Firstly, object of interest compensated for [6].
needs to be separated from other objects in the picture, Our solution uses the block-matching technique
and after that it needs to be tracked. This paper will that eliminates all aforementioned problems and
present some object tracking techniques that we were ensures real-time tracking with minimal hardware
considering, and some problems involved. resources. It was developed based on PII (700MHz)
The object tracking technique based on color is platform. The camera used was Logitech QuickCam
widely used technique [1] [2] [5]. The main problem is Express USB color web camera with 15 frames per
determining human skin color, if tracking object is second. Stepper motor is used for camera movement.
human, or determining the exact color of any other
object in general. HSI color space is used more often 2. COLOR-BASED OBJECT TRACKING
than RGB color space, because color is represented TECHNIQUE
only with H in HSI color space [3] [5]. If tracking
object is human, than usually the face is being tracked. This chapter will present our solution of object
H-value of the skin color is in the range from 0 to 50, tracking based on color.
or ranging from red to yellow color [2]. During that, In the first frame object was marked, and R, G and
certain problems arise with other parts of the body, B values of object pixel were read. In the second
such as arms, parts of the clothes or the surrounding frame, all pixels that are in the range [+20, -20] of R,
area that have the same color as the human face. On
G and B values of pixel from the first frame were At first, each of two successive frames, with even
separated. Considering that the object color lines from i-frame and odd lines from (i+1)-frame, are
significantly changes in the real situations because of combined. With this, parts of the picture that have
different light brightness of the area in which the changed due to the object movement are separated.
object is moving; it was necessary to update the color After that, pixels from newly combined picture are
from frame to frame. Updating is realized by separated using following expression:
calculating average R, G and B values of separated
pixels. In order to speed up the processing and to 2 ⋅ Y2 − DIL − ER > 50 , (1)
decrease the possibility of separation of similar
colored objects, the object was separated in the limited where:
area of the picture. The size of the limited area
depends on predicted object movement between two
DIL = max(Y1 , Y2 , Y3 ) , (2)
frames.
After object was separated, its motion between ER = min(Y1 , Y2 , Y3 ) . (3)
two frames is calculated as a difference between
Y1, Y2 and Y3 are the values of pixel’s luminance
centers of object’s circumscribed rectangle.
component for every 3 successive pixels in vertical
The use of HSI color space in the color based
line of the picture. Expression (1) is obtained
object tracking technique makes tracking more
experimentally.
efficient then when using RGB color space.
Even when we have applied aforementioned
limitations we could not perfectly separate object of
interest from other similar moving objects. Figure 1
shows good object separation when there are no other
unwanted similar objects in its vicinity, and Figure 2
shows bad object separation when there is unwanted
object in object’s vicinity. These two figures show
results gained by using HSI color space. a) b)

c) d)
Figure 1. Good object separation (there are no Figure 3. a) Frame i, b) frame (i+1), c)
unwanted objects in its vicinity). combined picture, d) separated area that
represents area of object movement.
Figure 3 shows the separation of object’s
movement area between two frames.
Object movement is determined like brunt’s
difference of separated pixels between two combined
pictures. Brunt is calculated upon general brunt
expression:
Figure 2. Bad object separation (there is
W 
unwanted object in its vicinity).
T X =  ∑ NWi ⋅ i  N , (4)
 i =1 
3. OBJECT TRACKING TECHNIQUE
BASED ON OBJECT MOTION  H 
BETWEEN TWO FRAMES TY =  ∑ NH i ⋅ i  N . (5)
 i =1 
This chapter will present our solution of object W and H are picture width and height, NWi and NHi
tracking based on object motion between two frames. are numbers of picture’s separated pixels in
column and row i, and N is overall number of picture’s separated it in limited area of other frames, and in that
separated pixels. way we have partially solved this problem. The size of
We were encountered two problems using this limited area depends on predicted object movement
technique. The first problem was separating object of between two frames. Second problem arises due to
interest from other objects in its vicinity that are not of camera movement. During the tracking, the camera
interest and which are also moving. We have marked moves and that creates the movement of the still area
the object in the first frame and after that we have that must be compensated for.

Figure 4. The position of the block, in relation to the tracked object in five frames.

Figure 5. The position of the block, in relation to the tracked object in five frames, when there are similar objects
in its vicinity

from frame to frame solves problem about human


4. OBJECT TRACKING BASED ON shape that changes during movement.
BLOCK-MATCHING TECHNIQUE Block’s comparison between successive frames is
done via summed absolute difference of pixel
With object tracking based on block-matching characteristics within blocks. As every pixel is
represented with three characteristics: blue (B), green
technique we have succeeded in eliminating all
aforementioned problems of other techniques. (G) and red (R), absolute difference is average
Block-matching technique is widely used in image difference of R, G and B values. The block that has the
lowest summed absolute difference is the best-
processing [7] [8]. Square block of specific size which
center is reference pixel, is compared with square matched block. Also, the summed differences in
subsequent frames must be lower then summed
blocks of same size in subsequent frame in the vicinity
of reference pixel from previously frame. Purpose of differences for no motion case by given threshold. In
block comparison is to find best matched blocks in two that way we have avoided cases when there is a
uniform area of the picture to miscalculate motion
successive frames.
In our solution we have used 40x40 pixel square direction of the object.
block that we have compared in 80x80 pixel area. When the best matched blocks are determined,
motion direction of object is the difference between
The area in which matching takes place is
predicted area of object movement between two centers of those two blocks.
frames. Figures 4 and 5 show positions of the blocks in
relation to the tracked object, with fixed camera,
In the first frame, block is centered to the object
that is being tracked, and in the second frame the most during five frames. Figure 5 shows that other similar
similar block to the block from the first frame is unwanted objects moving in the vicinity of the tracked
object do not interfere with the tracking.
searched and matched by block-matching technique. In
the third frame, the most similar block to the block In order to speed-up the image processing the
from the second frame is searched and matched by number of pixels inside the block, which we have used
to calculate the summed difference, was reduced. The
block-matching technique and etc. Block’s updating
number of referent pixels in the area (80x80 pixel) in
which the object moves was reduced 4 times.
Figure 6. Object tracking using block-matching technique while the camera moves in 10 frames.

The autonomous robot caddy is designed to be a


5. CAMERA CONTROL replacement for a human golf caddy. Robot caddy
main functions are: to carry golf clubs, to follow its
master after receiving command to move and to be still
For the camera motion we have used the standard
stepper motor (1.8º per step, unipolar) that moves the while its master plays.
camera on the horizontal axis from -180º to +180º. In order to apply our algorithm on robot caddy, it
is necessary that the robot is equipped with camera.
Based on the information of the object movement
obtained by block-matching technique, PC generates Robot follows its master using movement information
control signals to stepper motor driver that controls the obtained from our algorithm, supplied to robot’s
moving control algorithm.
camera. One signal controls the camera direction
movement (right or left), and the other signal controls Figure 7 shows robot caddy.
the number of steps that stepper motor makes. The
adjustment of the impulse and the pause of the second 7. EXPERIMENT
signal influences the speed of the motor. We have
achieved the smallest possible delay (the biggest Color-based object tracking technique

possible number of frames per second) by adjusting 300 True object movement -------
the impulse and the pause (pulse duty factor) of the Object tracking .......

second signal. 250

Figure 6 shows the object tracking using block-


matching technique while the camera moves. 200
x [pixel]

150
6. USAGE
100

50

0
0 100 200 300 400 500 600 700 800
Number of frames

(a)

Figure 7. Robot caddy.


Algorithm presented in this paper, among other
applications, can be applied on autonomous robot
caddy.
300
Error of color-based object tracking technique Figure 9. (a) Comparison of the true object
movement and object tracking based on object
250
motion between two frames, on the horizontal
axis, in the 850-frame sequence, (b) mean-
square error of object tracking technique based
Mean-square error [pixel]

200
on object motion between two frames, in the
150
850-frame sequence.
Object tracking based on block-matching technique

100
300 True object movement -------
Object tracking .......

50
250

0 200
0 100 200 300 400 500 600 700 800
Number of frames

x [pixel]
150
(b)
100
Figure 8. (a) Comparison of the true object
movement and color-based object tracking, on 50
the horizontal axis, in the 850-frame sequence,
(b) mean-square error of color-based object 0
tracking technique, in the 850-frame sequence. 0 100 200 300 400 500
Number of frames
600 700 800

Object tracking based on object motion between two frames (a)


Error of object tracking based on block-matching technique
300 True object movement -------
300
Object trcking .......

250
250

200
Mean-square error [pixel]

200
x [pixel]

150
150

100
100

50
50

0
0 100 200 300 400 500 600 700 800
Number of frames 0
0 100 200 300 400 500 600 700 800
(a) Nuber of frames

Error of object tracking technique based on motion between two frames (b)
300

Figure 10. (a) Comparison of the true object


250 movement and object tracking based on block-
matching, on the horizontal axis, in the 850-
Mean-square error [piksel]

200 frame sequence, (b) mean-square error of


object tracking based on block-matching
150 technique, in the 850-frame sequence.

100
All three techniques were tested in the same 850-
50
frame sequence, in order to show validity of them and
to compare them. In the frame sequence from 0 to 600
0
object is moving, and in its vicinity other similar
0 100 200 300 400 500
Nuber of frames
600 700 800
objects are also moving. In the frame sequence from
600 to 850 object is rotating.
(b)
Figures 8a, 9a and 10a show the curve of true
object movement and the curve of object tracking
obtained by certain technique, on the horizontal axis,
in the 850-frame sequence. As curves are more tracked is human, there is no need for object to be
matched tracking technique is better. Figures 8b, 9b face-turned toward camera.
and 10b show mean-square error of object tracking Algorithm can be upgraded, for example some
when object is being tracked with certain technique. preprocessing can be added in order to emphasize
Mean-square error contains error of object movement certain distinguishable characteristics of the object,
on both, horizontal and vertical axis. and which would in turn make the tracking more
From Figure 8 can be concluded that with efficient.
application of color-based tracking technique, object If the camera is mounted on a moving robot, the
can be tracked very efficiently when in object’s complete system could be autonomous.
vicinity there are no similar objects. Around 460th and
540th frame we have significant deviation between the
curve that represents true object movement and the 9. LITERATURE
curve that represents color-based object tracking. The
reason for this deviation is that in those frames in the [1] S. Lenser, J. Bruce and M. Veloso, “CMPack: A
vicinity of the object was unwanted object of similar Complete Software System for Autonomous Legged
characteristics. From the graph it can be seen that Soccer Robots,” in Proc. of the Fifth International
color-based tracking technique is not efficient during Conference on Autonomous Agents, Montreal, Canada,
the rotation of the object (the face is not turn toward May 2001.
the camera). [2] C. Garcia, G. Tziritas, “Face Detection Using Quantized
Figure 9 shows that the tracking technique based Skin Color Regions Merging and Wavelet Packet
on object motion between two frames is much more Analysis,” IEEE Trans. On Multimedia, 1(3), September
1999, p.264-277.
sensitive to movement of similar objects in its vicinity
[3] W. J. Hong, “Camera/Tracking Software
in comparison to color-based object tracking Documentation,” work done under the KOSEF/NSF
technique. The tracking technique based on object Summer Research Program at the Korea Advanced
motion between two frames handles the tracking of the Institute of Science and Technology (KAIST), August
rotation of the object more efficiently. 1998.
Figures 10 shows that the block-matching [4] Yu Zhong, Anil K. Jain, Marie-Pierre Dubuisson-Jolly,
technique is very efficient when there are similar “Object Tracking Using Deformable Templates,” IEEE
objects in the tracked object’s vicinity, but also it is Trans. on Pattern Analysis and Machine Intelligence,
very efficient if the tracked object is rotating. The 22(5): 544-549, 2000.
[5] V. Krüger, R. Herpers, K. Daniilidis and G. Sommer,
graph 10b shows that the mean-square error of the
“Teleconferencing using an attentive camera system, ”
block-matching technique is the smallest (mean-square Proc. of 2nd Int. Conf. on Audio- and Video-based
error contains error of object movement on both, Biometric Person Authentication, University of
horizontal and vertical axis). Maryland, USA, pp. 136-141, 1999.
[6] K. Daniilidis, Ch. Krauss, M. Hansen, and G. Sommer,
8. CONCLUSIONS AND FUTURE “Real-Time Tracking of Moving Objects with an Active
Camera,” Journal of Real Time Imaging, 4:3-20, 1998.
THOUGTHS [7] J. Lu and M. L. Liou, “A simple and efficient search
algorithm for block-matching motion estimation,” IEEE
This paper addresses camera object tracking Trans. Circuits Syst. Video Technol, vol. 7, no. 2, pp.
technique. Object is marked in the first frame, after 429-432, April 1997.
that it is separated and tracked by camera. Our [8] Jie-Bin Xu, Lai-Man Po and Chok-Kwan Cheung,
algorithm is able to track any moving object, and “Adaptive motion tracking block matching algorithms
because of that object must be marked in the first for video coding,” IEEE Trans. on Circuits and Systems
frame. for Video Technology, vol 97, pp. 1025-1029, October
1999.
Pixel’s characteristics that represent moving
object during tracking are updated from frame to
frame, and because of that our algorithm can be used
in wide brightness range of light in the area. In that
way, the tracking is possible both, in the daylight and
in the indoor light.
Due to the fact that the square block is updated
from frame to frame in tracking based on block-
matching technique, object that is being tracked can
change its shape (no need for object that is being
tracked to be rigid). In cases when object that is being

View publication stats

Potrebbero piacerti anche