ramya_Motion_Detection

International Journal of Engineering Research-Online
A Peer Reviewed International Journal
Articles available online http://www.ijoer.in
Vol.3., Issue.2, 2015
482 RAMYA R, SUDHAKARA B
I. INTRODUCTION
The moving object detection in video pictures has
attracted a great deal of interest in computer vision.
For object recognition, navigation systems and
surveillance systems, object detection is an
indispensable first step. Object detection has
significance in real time environment because it
enables several important applications such as
Security and surveillance [1] to recognize people, to
provide better sense of security using visual
information. Moving object detection is the basic
step for further analysis of video [11]. It handles
segmentation of moving objects from moving
background objects. This not only creates a focus of
attention for higher level processing but also
decreases computation time considerably.
Commonly used techniques for object detection
are background subtraction, statistical models,
temporal differencing and optical flow[10]. Due to
dynamic environmental conditions such as
illumination changes, shadows and waving tree
branches in the wind object segmentation is a
REVIEW ARTICLE ISSN: 2321-7758
MOTION DETECTION IN NON-STATIONARY BACKGROUND USING 'ORB' FEATURE
MATCHING AND AFFINE TRANSFORMATION FOR VIDEO SURVEILLANCE SYSTEMS
RAMYA R1
, SUDHAKARA B2
1
M.Tech Student, Department of Computer Science and Engineering, Srinivas Institute of Technology Mangaluru,
Karnataka, India
2
Assistant Professor, Department of Computer Science and Engineering, Srinivas Institute of Technology
Mangaluru, Karnataka, India
Article Received: 20/04/2015 Article Revised on:24/04/2015 Article Accepted on:28/04/2015
ABSTRACT
Visual surveillance systems have gained a lot of interest in the last few
years due to its importance in military application and security. Surveillance
cameras are installed in security sensitive areas such as banks, train stations,
highways, and borders. In computer vision, moving object detection and tracking
methods are the most important preliminary steps for higher-level video analysis
applications.
Moving objects in moving background are an important research area
of image-video processing and computer vision. Feature matching is at the base
of many computer vision problems, such as object recognition or structure from
motion. ORB is used for feature detection and tracking. The objective is to track
the moving objects in a moving video. Oriented Fast and Rotated Brief (ORB)
which is a combination of two major techniques: Features from Accelerated
Segment Test (FAST) and Binary Robust Independent Elementary Features
(BRIEF).Mismatched features between two frames are rejected by the proposed
method for a good accuracy of compensation. The Residues are removed using
Morphological Operation. The Frame Differencing methods are compared with
the proposed ORB feature matching algorithm to detect the accuracy and
efficiency.
Keywords—Object Detection, Visual Surveillance, motion detection.
©KY Publications

Vol.3., Issue.2, 2015
difficult and significant problem that needs to be
handled well for a robust visual surveillance system.
The goal is to detect moving objects in moving
background robustly with a real-time performance
using ORB feature matching. The main objective of
moving object detection aims at extracting moving
objects that are of interest in video sequences.
Fig.1. System Architecture
The overall architecture of the system is shown in
figure 1. The user stores pre-recorded videos
detecting the moving objects in moving background.
Object Detection is the identification of an object in
a video. The video is divided into many frames based
on the size of the video. Feature Detection is used to
find some interesting features in the image for
example corners, edges, blobs etc.ORB match
corresponding points across frames accurately. FAST
is a corner detection method, which could be used
to extract corner points. Similarity Transform is used
to detect the changes in the object shape due to the
changes in its structure. So morphology operations
are used to remove residues and ghost in the
individual frames.
literature survey
This section briefly outlines the related
works.The goal of background subtraction is to
remove the background in a scene by describing an
adequate model of the background [2]. The result is
that only interesting objects are left in the scene for
tracking and further analysis. This technique
generally has a low computational cost. It is done in
a pixel by pixel fashion. However, in traditional
background subtractions are susceptible to
environmental changes, for example, in the cases of
gradual or sudden illumination change. These
changes alter the background model. The result of
background subtraction is always contaminated by a
large number of erroneous foreground pixels.
However, one drawback is that it is vulnerable to
scene dynamics and clutter. It works only for static
background and dynamic background model update
is required for dynamic background scenes [3].
Color Histograms are used for object tracking
because they are flexible in the type of object that
they can be used to track, including vehicles and
people. A single feature does not provide enough
information about the object being tracked. The
problems arise when target object and background
have similar color distribution. It may be difficult to
distinguish the object from background .Higher level
feature descriptors can also be used for object
tracking [4].
The method of Frame difference is used to
detect moving objects. The method does not have
background model. The current frame is simply
subtracted from the previous frame and if the
difference in pixel values is greater than threshold,
the pixel is considered as part of the foreground
[12]. The objects must be continuously moving in
each frame. If the object does not move for more
than frame period it becomes the part of
background pixel. Difficult to determine the value of
the threshold [5].
Kalman filter Object tracking problem can be
formulated as a state estimation problem given
available observation. Kalman filtering is popularly
used for object tracking because it has been shown
to be very successful for linear and Gaussian
dynamic state estimation problems and is still very
reliable in cases like clutter and occlusions. Kalman
filter segments moving objects by cluster the image
features, which will be wrong if the image features
are some mismatch [6].
ORB IMPLEMENTATION DETAILS
A. FAST Keypoint Orientation
ORB [7] can be used to detect local keypoints in an
image with good performance and low
computational cost. In order to extract keypoint fast
and accurate, ORB start by detecting Features from
Accelerated Segment Test (FAST) points in the
image. The FAST [8] segment test criterion operates
by considering a circle of sixteen pixels around the
corner candidate p. The detector classifies p as a
corner if there exists
a set of 12 contiguous pixels in the circle which are
all righter

Vol.3., Issue.2, 2015
than the intensity of the candidate pixel I(p) plus a
threshold t,
or all darker than I(p)-t. The test examines only the
four compass directions for a high speed. If p is a
corner then at least three of these must all be
brighter than I(p)+t or darker than I(p)+t. For
rotation invariant, the keypoints orientation is
computed by the moments of keypoint’s patch:
I(x, y) is pixel’s intensity at position x and y of the
keypoint’s patch.
B. Rotation-Aware Brief
It is needed to describe each keypoint for a good
match. The descriptor of ORB is the improvement of
Binary Robust Independent Elementary Features
(BRIEF). BRIEF is a recent feature descriptor that use
simple binary test between pixels in a smoothed
image patch [9]. Consider a smoothed image patch,
p. A binary test τ is defined by:
where p(x) is the intensity of p at a point x, the
feature is defined as a vector of n binary test:
The test pairs of x and y are selected by PCA for a
good discriminative.
OUR IMPROVED MATCHING METHOD FOR
DETECTING MOVING OBJECTS
A. Extracting and matching ORB feature
In existing motion compensation methods, the main
part of computation time is occupied by features
extracting and matching. ORB is an order of
magnitude faster than SURF, and over two orders
faster than SIFT. So it is reasonable to choose ORB
features for a faster computation to improve the
motion compensation methods. In addition, ORB
descriptor is binary strings. Its similarity is evaluated
more efficiently by using the Hamming distance than
2 L norm which SIFT and SURF are using.
B. Rejecting the mismatched pairs
Although the descriptors in the previous frame can
be matched with the next frame robustly by brute-
force method, some mismatched descriptors are still
existed in traditional matching method. These
mismatched descriptor pairs would damage the
estimation results and lead detection failed, as show
in SIFT based and SURF based method. So we
proposed a method of distance constraint method
to reject mismatched descriptor pairs. We know that
even if the camera would move fast, the motion
distance between two adjacent frames wouldn’t be
very large. That means if a keypoint located at (x, y)
in the previous frame. The matched keypoint would
be in the neighbourhood of the same position in the
next frame. We assume that this distance is less
than d. If the distance between two descriptors
positions of the matched pairs is larger than d, this
descriptors should be rejected for good matching
results.
Fig. 2. Matched pairs of ORB feature between two
frames.
Fig 3. Retained pairs after rejecting mismatched
pairs by distance constraint.
C. Remove ghost and residues
The motion model of a camera can be described by
an affine transformation, so the transformation of
(k-1)-th frame point [x y]
T
to the (k+1)-th frame point
[u v]
T
can be written as
m1 m6 the motion parameters. The equation
above can be rewritten as:

Vol.3., Issue.2, 2015
Let (x, y) and (u, v) be the descriptor position of
matched pairs, the parameters can be solved easily
by the least squares method. The previous frame is
transformed by the affine transformation of
parameters we have estimated, the motion between
these two frames can be compensated. Then the
Frame difference method is used to detect moving
objects. If the objects are moving very fast in the
video, then a large part of moving objects area can
be detected which will make the detection result
more perfectly.
Most of existing method ends at this step, however,
residues and ghost would be still appeared in the
resulting image through precise compensation, such
as SIFT or SURF based method. Part of this problem
is because of dynamic background (for instance,
swing tree leaves or changing light). We proposed
that remove residues and ghost by Logic AND
operation between two result frames after Frame
difference.
V. MOVING OBJECT SEGMENTATION
Moving objects detection in moving background has
been introduced above. Entire algorithm is
summarized as follows:
1. ORB features are extracted and matched between
two frames.
2. Mismatched descriptor pairs are rejected by the
method of distance constraint.
3. Parameters of affine transformation are
computed precisely by the rest of matched pairs.
4. The (k-1)-th frame Ik-1 is transformed to I
1
k-1 by the
affine transformation.
5. I
1
k-1 subtracts Ik-1 for Frame difference:
6. Segment moving objects after Logic AND
operation between two subtracted images and
morphology method.
Where O(x, y) is resulting image after detecting
moving objects. 255 means moving region, 0 means
static region.
SIMULATION AND RESULTS
Experimental results of the proposed method are
presented in this section. The video sequences are
processed using OpenCV in MATLAB. The Frame
Differencing methods are compared with the
proposed ORB feature matching algorithm to detect
the accuracy and efficiency. The Figure 4 represents
Motion pixels truly detected as motion.
Figure 4: True Positive
The Figure 5 represents Background pixels truly
detected as background.
Figure 5: True Negative
Figure 6: False Positive
The Figure 6 represents Background pixels falsely
detected as motion.
Figure 7: False Negative
The Figure 7 represents Motion pixels falsely
detected as background.

Vol.3., Issue.2, 2015
CONCLUSION
In this paper, a real-time video of moving object
detection in moving background got by moving
camera based on ORB feature matching is
presented. The motion can be compensated
accurately and quickly after rejecting mismatched
descriptor pairs. The camera motion can be
compensated accurately by rejecting mismatched
descriptor pairs, because of the using of ORB
features, the motion can be compensated accurately
and quickly after rejecting mismatched descriptor
pairs. Compared with other motion detection
algorithms, the proposed method is much better to
detect the moving objects.
REFERENCES
[1]. Songyun Xie, Wanpeng Zhang, Wang Ying,
and Khalid Zakim "Fast Detecting Moving
Objects in Moving Background using ORB
Feature Matching" International
Conference on Intelligent Control and
Information Processing (ICICIP) 2013.
[2]. Ka Ki Ng and Edward J. Delp, Object
Tracking Initialization using Automatic
moving Object Detection, Proceedings of
the SPIE Conference on Visual 7
Information Processing and
Communication, vol.7543, San Jose, CA,
January 2010.
[3]. Samera. J. H. Jean and F. L. Lian, "Robust
Visual Servo Control of Mobile Robot for
Object Tracking Using Shape Parameter,"
Control Systems Technology, IEEE
Transactions on, vol. 20, pp. 1461-1472,
2012.
[4]. B M. Teutsch and W. Kruger, "Detection,
Segmentation, and Tracking of Moving
Objects in UAV Videos," in Advanced Video
and Signal-Based Surveillance (AVSS), 2012
IEEE Ninth International Conference on,
2012, pp. 313-318.
[5]. H. Ying, X. qiu, J.Song and X.Ren, Particle
filtering object trackingbased on texture
and color, Proceedings of the IEEE
International Symposium on Intelligence
Information Processing and Trusted
Computing, Huanggang ,China, October
2010,pp. 62 6630.
[6]. I. Haritaoglu, D.Harwood, and L.S. Davis,
W4: Real-time surveillance of people and
their activities,IEEE Transactions on Pattern
Analysis and Machine Intelligence, vol. 22,
no.8. Washington, DC,USA:IEEE Computer
Society, August 2000,PP. 809830.
[7]. E. Rublee, V. Rabaud, K. Konolige, and G.
Bradski, "ORB: an efficient alternative to
SIFT or SURF," in Computer Vision (ICCV),
2011 E International Conference on,
2011, pp. 2564-2571.
[8]. E. Rosten, R. Porter, and T. Drummond,
"Faster and better: A machine learning
approach to corner detection, “Pattern
Analysis and Machine Intelligence, IEEE
Transactions on, vol. 32, pp. 105-119,2010.
[9]. M. Calonder, V. Lepetit, C. Strecha, and P.
Fua, "Brief: Binary robust independent
elementary features," Computer Vision–
ECCV 2010, pp. 778-792, 2010
[10]. Mr. Mahesh C. Pawaskar, Mr. N.
S.Narkhede and Mr. Saurabh S. Athalye,”
Detection Of Moving Object Based On
Background Subtraction”, 2014
[11]. Rafael C. Gonzalez, Richard E. Woods
“Digital Image Processing” Pearson
Education South Asia Ltd pp 630-639
[12]. M. Kalpana Chowdary, S.Suparshya Babu,
S.Susrutha Babu, Dr.Habibulla Khan “FPGA
Implementation of Moving Object
Detection in Frames by Using Background
Subtraction Algorithm” International
conference on Communication and Signal
Processing, April 3-5, 2013, India ©2013
IEEE pp 1032-1036.

ramya_Motion_Detection

More Related Content

What's hot (18)

Viewers also liked (12)

Similar to ramya_Motion_Detection (20)

More from ramya1591 (7)

Recently uploaded (20)

ramya_Motion_Detection