Anda di halaman 1dari 2

IJSRD - International Journal for Scientific Research & Development| Vol.

4, Issue 05, 2016 | ISSN (online): 2321-0613

Background Modeling and Moving Object Detection


Miss. Neha S. Patil1 Dr. S. R. Gengaje2
1
Student 2Head of Dept.
1,2
Department of Electronics Engineering
1,2
WIT, Solapur, India
Abstract The most challenging task for most object foreground pixels will also be absorbed into the background
detection applications is to accurately segment foreground model which results in a high false negative rate. This is
from the background under the presence of heavy also known as the foreground aperture problem. Also in
illumination changes. Moving object detection provides the areas, such as building gates or highways where people
separation of pixels in the foreground and background. While walk, it is most difficult to get good background models
classifying, it is most important to model background for without filtering out the effects of foreground objects. A
better results as background may contain some inappropriate variety of methods have been proposed for moving object
data like flickering monitors, swaying trees etc. The detection and many different features are utilized for
proposed work uses a texture based method i.e. Local binary modeling background. This paper proposes an approach that
pattern for modeling the background and detecting moving uses discriminative texture features to capture background
objects. This approach is very robust in terms of gray scale statistics. Here LBP operator is used to which has shown
variations, as the LBP operator is invariant to monotonic excellent performance in many applications.
transformation of gray scale. Major steps involved in Texture classification is an active research topic in
proposed system are background modeling, background computer vision. Kashyap and Khotanzad were among the
subtraction and then foreground detection. The proposed first researchers to study rotation invariant texture
work is implemented on MATLAB platform. classification by using circular autoregressive model [1].
Key words: Modeling Object Detection, Moving Object LBP has also been adapted to many other applications, such
Detection as face recognition[2], dynamic texture recognition [3] and
shape localization [4].LBP has its own properties i.e. it has
I. INTRODUCTION tolerance against illumination changes and it is
Basically, object detection is a computer technology which computationally very simple. To make LBP more suitable
is related to computer vision and image processing which for real world scenes it has been modified and this modified
deals with detecting single occurrence of semantic objects of version is used in the proposed method. LBP has a
a particular class like humans, buildings, cars etc. disadvantage in handling moving shadows. And this is the
Background subtraction and moving object detection has difficult problem to solve with the background modeling. A
several uses in various fields like traffic monitoring, human survey of moving shadow detection is presented in [5].
action recognition and also many other applications like
digital forensics. Basically, background subtraction means II. PROBLEM STATEMENT
subtracting the current image from reference background The main objective of the proposed work is to design a
model. framework for modeling the background using texture
information operator detecting moving objects in a video
sequence.

III. METHODOLOGY
The input will be videos downloaded from internet or
recorded videos from CCTV camera. All videos will be in
.avi format for simplification. Also the frame rate will be
15-20 fps. Input video is first converted into the frames.
After that LBP operator is applied on every frame for
background modeling and then foreground is detected by
Fig. 1: Generalized Block Diagram of Object Detection comparing histograms of the model frame and current
There are some challenges in background modeling frame.
like swaying trees, weaving water or flickering monitors and
also illumination changes. Various techniques have been
introduced for background modeling. Each having strength
and weaknesses. These background modeling techniques
can be classified as basic background modeling, statistical
background modeling, fuzzy background modeling. Other
classifications also can be found in terms of prediction. The
simplest background model assumes that the intensity values
of the pixel can be modeled by a single unimodal
distribution. However, a single model cannot handle
multiple backgrounds like weaving trees. Again, if the Fig. 2: Block diagram of proposed system
model is adapting too quickly then slowly moving This block diagram consists of the following steps:

All rights reserved by www.ijsrd.com 392


Background Modeling and Moving Object Detection
(IJSRD/Vol. 4/Issue 05/2016/097)

Background modeling and false negatives. The number of foreground pixels which
Foreground detection are missed are called as the false negatives and the
Object detection background pixels that are marked as foreground are called
as the false positives.
A. Background Modelling
Background modeling is the core component of any object
detection technique. The background model should be
adaptable to changes in background like illumination or
weaving trees etc. Proposed system uses a texture based
method for background modeling i.e. local binary pattern. It
has several properties that favor its usage in background
modeling. It is invariant to the monotonic variations of gray
scale. Also there is no need to normalize the input image if
we use LBP. This operator labels each pixel of an image
region by thresholding the neighborhood of it with the
centre value and considering the result as a binary number.
LBP is calculated as,
p1

LBP(P,R) (xc , yc ) = s(g p g c )2p Fig. 3: Result of Object Detection


p=0
Condition, V. CONCLUSION
s(x) = 1 for x 0 and s(x) = 0 for x < 0 A method for moving object detection is presented, which
LBP histogram is calculated for every block of new uses a texture based LBP operator for background modeling.
video frame. After this the new block histogram say Xt is As it as invariant to monotonic gray scale changes, so the
compared against the k model histograms by using the method gives good results in illumination changes occurred
histogram intersection as the distance measure. If none of in natural scenes.
the histogram is close enough to the new histogram, then
model histogram with lowest weight is replaced with the
ACKNOWLEDGEMENT
new histogram and low initial weight is given to it. And if
model histogram is close enough to the new histogram is I would like to express my feeling of gratitude towards the
found, then bins of the histogram are updated. And based on number of people who were supportive in successfully
these histograms we have to calculate the foreground and making the project. I am also thankful to our project
background pixels. Histogram intersection is done to find coordinator and project guide Dr. S.R. Gengaje sir for
out the common part in two frames. giving valuable suggestions, technical guidance during the
project work that I had undertaken. I am thankful to our
B. Foreground Detection principal Dr. halkude for the constant encouragement and
Foreground detection is done by comparing input frame assistance they provided me.
with the current background model. The areas of the input
frame which do not fit to the background model are REFERENCES
considered as foreground. So, here new block of histogram [1] Marko heikkila and matti pietikainen,A texture based
Xt is compared against existing B background histograms method for modeling the background and detecting
which are selected at the previous time instant. If match is moving objectsIEEE transactions on pattern analysis
not found, then the block is considered as concerned to the and machine intelligence, vol.28 no.4, April 2006.
background. Otherwise it is considered as belonging to the [2] C. stauffer and W.E.L Grimson,Adaptive background
foreground. mixture models for real time tracking,proc. IEEE
C. Object Detection CS.conf. computer vision and pattern
After marking the background and foreground we will recognition,vol.2,pp.246-252,1999.
simply get the moving object detected. Figure 2 shows the [3] T. Matsuyama, T. Ohya, and H. Habe, Background
output of proposed system. The histogram of every frame is Subtraction for Non- Stationary Scenes, Proc. Asian
also calculated for each frame. Conf. Computer Vision, pp. 622-667, 2000.
[4] A. Monnet, A. Mittal, N. Paragios, and R. Visvanathan,
Background Modeling and Subtraction of Dynamic
IV. RESULTS AND DISCUSSIONS
Scenes, Proc. IEEE Intl Conf. Computer Vision, vol.
Below figure shows the output of moving object detection 2, pp. 1305-1312, 2003.
using texture based LBP operator. This method belongs to [5] Q. Zang and R. Klette, Robust Background
nonparametric methods i.e. no assumptions are needed about Subtraction and Maintenance, Proc. Intl Conf. Pattern
the underlying distributions. Recognition, vol. 2, pp. 90-93, 2004.
While observing the figure we can see that there [6] T. Ojala, M. Pietikainen, and D. Harwood, A
are some pixels related to foreground are not detected Comparative Study of Texture Measures with
properly. And also we can see that there are some pixels Classification Based on Feature Distributions, Pattern
related to the foreground are missing. So the numerical Recognition, vol. 29, pp. 51-59, 1996.
evaluation of this situation is done in terms of false positives

All rights reserved by www.ijsrd.com 393