object ttacking real time embdded ystem using imag processing

Object Tracking In Real Time Embedded System
Using Image Processing
M. S. Chelva Dr. S.V. Halse Dr.B.K.Ratha
Research Scholar, SRTM University Karnataka States Women’s University Utkal University
Nanded, Maharashtra, India Vijaypur, Karnataka, India Bhubaneswar,Odisha,India
mschelva@gmail.com drsvhalse@rediffmail.com b_ratha@hotmail.com
Abstrac- Robotics now a days becoming more popular due to diverse applications. Vision to a robot has widened the range of
applications and is developing rapidly. Robots with the power of image processing are used for object identification, surveillance and
navigation. Vision applications generally deal with finding a part and orienting it for robotic handling or inspection before an
application is performed. In this paper we developed algorithm & implemented on robotic system having embedded processor, process
the image that detects & track the object in real time. The results are promising.
Key words: Object tracking, Real time, ARM processor, Median Filter, Image processing
I.INTRODUCTION
The problem of object detection can be seen as a classification
problem, where a need to distinguish between the object of
interest and other background objects. With the advent of
System On-Chip (SOC) technology it is now possible to
integrate complex hardware functionality on a single chip. For
such applications, the software part is usually the most
performance critical factor. For this reason it is important that
such components are highly optimized in terms of speed while
using a relatively small section of the available embedded
memory. The object recognition and manipulation algorithms
are characterized by their computational complexity due to the
size of both the image and the source system added to the
large number of complex arithmetic operations. It is extremely
desirable that such applications are performed on a standard
SOC embedded processor without the need for large and
expensive memories and co−processors. The main aim of this
paper is therefore to show the feasibility of implementing
object recognition & tracking algorithms on a standard SOC
target which provides the advantage of flexibility and speed.
The model used in image processing applications to detect
object on basis of its appearance can be broadly classified as
(a) model-based technique,[4] in which the model of
predefined in the system is compares with the current image
frame to determine the presence of object. This may include
features like lines curves etc. (b) appearance-based technique,
wherein the variation in the appearance of object are used to
train the object models; [4] (c) local appearance technique
uses small part of an object to train the object models. [4]
II.LITERATURE REVIEW
Enhanced image detection on an ARM based embedded
system [3] presents a new technique for the detection of ICs
within images of PCBs which uses region growing. The
application of region growing to the detection of ICs is a novel
approach. This limitation is the fact that only ICs within a
restricted band of grey scale intensities are detected. The
Hough transform is of order O(n6) compared to region
growing and find IC algorithms which have complexity O(n).
It has presented the porting of the techniques to an ARM7
System on-chip target. The porting of an object recognition
operator to this target system is novel and the feasibility of this
approach has been shown. In a paper on real time image
processing algorithms a suite of algorithms for determining
the possible trajectories of an autonomous robot while
navigating through obstacles is elaborated. A paper on
Estimating and Reducing the Memory Requirements of Signal
Processing Codes for Embedded Systems is helpful in
reducing memory footprint of the application.
The work gives overview of two embedded systems for
object detection and pose estimation using point features [1].
The feature detection step of the “Speeded-up Robust
Features” algorithm is increased by a special IP core. The first
system is used to perform object detection and is implemented
on Virtex-5 FPGA. The second system consists of an ARM-
based microcontroller and intelligent FPGA-based cameras
which support the main system [1].
The median filter is very widely used in image pre-
processing to enhance or smoothen the image. If the situation
demands to simultaneously reduce noise and preserve edges,
an effective option to convolution operation in image
processing is median filtering [7]. Order static (OS) filter is a
generalized form of median filter. The median filter traverses
Bot
Image Frame
Object
Distance (d)
Figure 1: Bot Tracking

the complete image frame pixel by pixel, modifies each value
by the median value of neighboring pixels. The window size
specifies the neighbors considered in filtering process and this
window slides, pixel by pixel over the entire image. The
modified value or the new value is the median of the sorted
pixel values that are present within the window, and then the
pixel is replaced by the new value or the median.
The other popular method for image pre-processing is the
morphological operations like opening and closing[8]. The
gray-scale morphological operations are used to eliminate
background from image frame and further the motion analysis
is performed [8]. The operation of algorithm is limited by the
speed of moving object. A watershed algorithm is used for
image segmentation to detect the object in image with low
contrast [9]. The difference image is used to remove noise and
then watershed transformation is applied to it.
"Image Processing Algorithm for Robotics on Embedded
System" [2] describes image processing implementation on
embedded platform for embedded application 3-D
environment. It is a real time operating system (RTOS) which
will run the Digital Image Processing Algorithms. These
algorithms are used to extract the information from the
images. The camera connected on USB bus captures images
on the ARM9 core running RTOS.
Depending upon the information extracted, the further process
is carried out. The camera is a simple CMOS USB-camera
module with a resolution of 0.3MP. Video4Linux API’s
provided by kernel are used to capture the image, and after
decoding, the object location is detected using image
processing algorithms. The embedded Linux kernel supports
multitasking feature and ensures that the task is performed
within the real time constraints. The OS makes system flexible
such as interfacing new devices, handling the system and
storage of data. ARM9 meets hard real time constraints that
are imposed in embedded environments, with low power and
performance trade-offs. Object detection applications are
associated with real-time performance constraints that
originate from the embedded system [2]. Embedded system
using the ARM 32-bit micro-controller has the feature of
image or video processing by using the variety of features and
classification algorithms have been proposed for object
detection.
III.METHODOLOGY
a. Algorithm
The implementation of an image processing algorithm on
embedded system is a critical task due to complexity in
implementation. As a standalone system, various factors like
memory, power consumption, space, etc. has to be considered
while selecting the platform and implementing the image
processing algorithm. Here, an algorithm to detect object is
implemented on the embedded processor board (EP9302) with
embedded-OS ported on it. The algorithm flowchart is shown
in Fig.1.
The main purpose of this system is to detect and track the
object in the real time world and determine direction in which
the robot should move so as to reach near the object. To detect
the object a USB camera is used which is interfaced to
processor. The output of camera is in compressed format
which is decoded by the processor and stored in memory. As
the image frame is in color format (RGB format), it requires
more processing time therefore, the complete frame is
converted to 8-bit grayscale image. The median filter is the
crux of this implementation. The object of interest is bigger in
size then the other smaller objects are assumed to be as noise
present in the image frame. As the property of median or
averaging filter is that it removes noise from the image or in
other words smoothens the image and detect the object.
Initially, the bot is at a certain maximum distance, defined
by the application. When distance (d) between bot and target
is maximum, the object would appear to be smaller in size as
compared to complete image frame captured by the camera.
In other words, the pixels covered by the object in the image
frame would be less and therefore the median filter of
smaller size is used to ensure that after smoothening the
image only the object is clearly seen in the image. The filter
size is decided by the specifications of application. Now, as
the robot system (Bot) advances toward the object, the object
would appear to be larger i.e. it covers more and more
number of pixel in the image frame and hence the filter size
is increased as the bot advances forward. After filtering, as
only the object of interest is present in the image frame,
correlation is used to get the approximate centroid of the
object.
b. Flowchart
The flowchart shown in Fig.2 shows the process of image
pre-processing and foreground object detection.

The pseudo-code of the implementation is as follows:
i. Initialize processor
ii. Configure camera to provide output
iii. Store object mask in SDRAM
iv. while(1)
malloc(input buffer)
Input buffer = click image(camera device ID)
Gray image = Gray-scale image of input image
new image = Median Filtering(gray image,
FilterSize)
Neg image = 255 – new image
r = rows of Neg image
c = pixels in one row
for i = 1 to r
for j = 1 to c
get correlation between image and mask with mask
center as (i, j)
if (correlation > threshold)
object present = true
break;
else
object present = false
end if
end j loop
if (object present == true)
object location = (i , j)
break;
end if
end iloop
free(input buffer)
free(gray image)
free(new image)
free(Neg image))
end while
IV.RESULT
The results of the image pre-processing algorithm on the
embedded processor are shown here. The Fig.3a is the input
image captured from the camera and Fig.3b is negative of the
filtered output image. It is seen that the noise in input image is
removed and the object of interest appears to be completely
segmented from the background in Fig.3b can be easily
extracted. Fig.4 shows the result of use of median filter on an
image with salt and pepper noise. The Fig.4b is generated n
MALAB image and the median filtering is carried out on
embedded processor. The Fig.5 shows the processed image at
every stage. Fig.5a is the gray scale input image of resolution
640x480. Fig.5b is the image filtered with mask of size 11x11
and Fig.5c shows the image after negation.
Figure 3: (a) Input Image (b) Filtered Output
The object location is detected by determining the count of
bright pixels within a specified size of matrix. If the count of
START
Capture Image
Decode and Convert image to
raw grayscale
Perform Median Filtering
Is Object
Present?
Drive Robot
Negate Image
No
Yes
Figure 2: Flowchart

bright pixel exceeds a defined threshold value, then the object
is present. The threshold value is dependent on the distance of
object from the camera.
(a) (b) (c)
Figure 4: Smoothen by processor -5x5 mask
(a) (b) (c)
Figure 5: (a) Input Image (b) Filtered image with 11x11 ask
size (c) Foreground object image
V.CONCLUSION
This paper presents the study of various techniques used to
detecting & tracking of an object in embedded systems. The
paper discussed in a way to detect & track the object in real
time image frame on embedded processor. The order-state
filter of fixed size is used as pre-processing image technique
that makes a clear distinction between the foreground and the
background. As the size of filter increases the efficiency of
detection of the object increases, however the processing time
of the system also increases. Obviously higher size filters are
not preferred in real time systems. The developed algorithm
provides good result on the specific real time system. The
algorithm is restricted to detect objects of size greater than the
size of background noise. The supervised algorithms can be
used to overcome this limitation. This complete system can be
used in the various security and surveillance applications.
References
[1]Michael Schaeferling, Ulrich Hornung, and Gundolf Kiefer,
“Object Recognition and Pose Estimation on Embedded
Hardware: SURF-Based System Designs Accelerated by
FPGA Logic,” International Journal of Reconfigurable
Computing, Volume 2012, Article ID 368351, 16 pages
doi:10.1155/2012/368351.
[2]Sudhir D. Zaware, Prajwal G. Awade, Chinmay A. Joshi,
R. V. Tornekar, " Image Processing Algorithm for Robotics on
Embedded System," International Journal of Industrial
Electronics and Electrical Engineering, ISSN: 2347-6982
Volume-2, Issue-12, Dec.-2014
[3]Evans, Jonathan R., and Tughrul Arslan. "Enhanced image
detection on an ARM based embedded system." Design
automation for embedded systems 6.4 (2002): 477-487.
[4]Parks, Donovan H. Object detection and analysis using
coherency filtering. Diss. McGill University, 2006.
[5]Rao, G. Mallikarjuna, and Ch Satyanarayana. "Object
Tracking System Using Approximate Median Filter, Kalman
Filter and Dynamic Template Matching." International Journal
of Intelligent Systems and Applications 6.5 (2014): 83.
[6]Hallinan, Christopher. Embedded Linux primer: a
practical, real-world approach. Pearson Education India, 2007.
[7]Srivastava, Hari Babu. "Image pre-processing algorithms
for detection of small/point airborne targets." Defence Science
Journal 59.2 (2009): 166.

[8]Dong, Nengli, et al. "New approach to detect dim moving
point targets based on motion analysis." International
Symposium on Optical Science and Technology. International
Society for Optics and Photonics, 2001.
[9]Hsieh, Feng-Yang, et al. "A novel approach to noise
removal and detection of small objects with low
contrast." Engineering, WIAMIS (2004).
[10]Kligys, Skirmantas, Boris Rozovsky, and Alexander
Tartakovsky. Detection Algorithms and Track Before Detect
Architecutre Based on Nonlinear Filtering for Infrared Search
and Track Systems. No. CAMS-98.9. 1. UNIVERSITY OF
SOUTHERN CALIFORNIA LOS ANGELES CENTER FOR
APPLIED MATHEMATICAL SCIENCES, 1998.
[11]Gonzalez, Rafael C., Richard E. Woods, and Steven L.
Eddins. Digital image processing using MATLAB. Vol. 2.
Knoxville: Gatesmark Publishing, 2009
[12]Boyle, Roger D., and Richard C. Thomas. Computer
vision: A first course. Blackwell Scientific Publications, Ltd.,
1988.
[13]LTD, ARM. "ARM920T technical reference manual."
ARM Ltd.
[14]EP9302 Reference manual.

object ttacking real time embdded ystem using imag processing

More Related Content

Similar to object ttacking real time embdded ystem using imag processing (20)

More from archanadeiva (8)

Recently uploaded (20)

object ttacking real time embdded ystem using imag processing