Applying edge density based region growing with frame difference for detecting moving objects in video surveillance systems

IJRET: International Journal of Research in Engineering and Technology eISSN: 2319-1163 | pISSN: 2321-7308
__________________________________________________________________________________________
Volume: 03 Issue: 04 | Apr-2014, Available @ http://www.ijret.org 693
APPLYING EDGE DENSITY BASED REGION GROWING WITH FRAME
DIFFERENCE FOR DETECTING MOVING OBJECTS IN VIDEO
SURVEILLANCE SYSTEMS
Abhilash K. Sonara1
, Pinky J. Brahmbhatt2
, Manoj D. Chaudhary3
1
M.E Student, ECE Department, L.D. College of Engineering, Gujarat, India
2
Associate Professor, ECE Department, L.D. College of Engineering, Gujarat, India
3
M.E Student, ECE Department, L.D. College of Engineering, Gujarat, India
Abstract
Detecting moving object is the key step in Video based Surveillance Systems. This paper presents a unique approach for detecting
moving object from a video sequence based on edge detection and Edge Density Based Region Growing. To start with the video
sequence is sub-divided into its constituent frames. The method proceeds by considering two consecutive frames at time intervals (t-1)
and t. Then after we use Sobel’s operator to detect the edges present in respective frames. Detecting edges helps in estimating the
object boundaries. After this we perform frame difference which eliminates most of the edges belonging to background. To highlight
the regions containing the moving object we apply Edge Density Based Region Growing algorithm on the edge difference image. At
last using Connected Component Analysis and Morphological Opening we extract the exact moving object. Experimental results
show that the method is robust enough to detect the moving objects with complex backgrounds.
Keywords: Contrast Stretching, Edge Density Based Region Growing, Edge Detection, Frame Difference, Morphological
Opening, Moving Object Detection, Video Surveillance.
-----------------------------------------------------------------------***-----------------------------------------------------------------------
1. INTRODUCTION
Moving object detection is the basic step for further
processing and analysis of video captured using surveillance
cameras. The advancement in computing power, availability
of large-capacity storage devices and high speed network
infrastructure has paved the way for cheaper, multi sensor
video Surveillance Systems. Video surveillance demands
design and implementation of intelligent systems that can
automatically detect, tack and recognize objects in the
captured video sequence. Moving object detection has a wide
number of applications including Visual surveillance, Event
detection; Activity based Human recognition, Path detection,
Object tracking etc [1]-[3]. Traditionally, the video outputs are
processed online by human operators and are usually saved to
tapes for later use only after a forensic event. After object
detection comes object classification step which is used to
categorize detected objects into predefined classes such as
human, vehicle, animal, clutter, etc. It is necessary to
distinguish objects from each other in order to track and
analyze their actions reliably. Based on the movements of
objects and background, video sequences can be categorized
into two types: moving objects with moving background and
moving objects with fixed background. Conventional
approaches to moving object detection include Frame
Differencing; Background Subtraction, and Optical Flow.
Background subtraction attempts to detect moving object by
subtracting the current image pixel-by-pixel from a reference
image (background image) that is created by averaging images
over time in an initialization period. The pixels where the
difference is above a threshold are classified as foreground
and the remaining serve as those belonging to background.
Though background subtraction techniques perform well at
extracting most of the relevant pixels belonging to moving
regions, they are usually sensitive to sudden dynamic changes
[4]. Frame differencing attempts to detect moving regions by
taking pixel-by-pixel difference over two or three consecutive
frames from the video sequence. This method is highly
adaptive to dynamic changes [5]. Optical flow methods
employ flow vectors of moving object over time to detect
moving regions in an image. Using these methods one can
detect objects in motion even from a moving camera [6]. The
only difficulty with optical flow methods is that they are
computationally complex and their real time implementation
requires specialized hardware.
The rest of the paper is organized as follows. Section 2 gives
brief overview of few existing methods for object detection.
Section 3 describes the proposed methodology in detail.
Experimental results are discussed in section 4 while section 5
concludes the paper by presenting scope for future work.

__________________________________________________________________________________________
2. LITERATURE SURVEY
Reference [7] utilizes textural edges in a frame for detecting
the moving object. The edges extracted edges from each frame
are represented as segments. Then after the authors have used
Local Directional Pattern (LDP) to represent the texture
information by considering the edge orientations in eight
different directions. In reference [8] the authors M.Sankari and
C. Meena have proposed a moving object detection
methodology that makes use of background subtraction based
on some statistical assumptions from the previous frame. The
authors H. S Mohana, Aswatha Kumar, and Shiva Kumar have
implemented a robust model-based tracker and classifier that
utilizes Kalman Filtering in [9].
3. METHODOLOGY
To detect the moving object in the surveillance video captured
from stationary camera the simplest method is frame
difference. Frame differencing has an advantage of low
computational complexity and ease of hardware
implementation. In this method the unchanged part is
eliminated during subtraction. In this paper we propose an
improved and effective method for object detection utilizing
framed difference and edge density map. The method is
further divided in two stages: 1) Pre-Processing and 2) Exact
Methodology.
3.1 Pre-Processing
This stage involves three steps: 1) Separating captured video
into its constituent frames 2) Transforming RGB frames to
grayscale images and 3) Contrast Stretching. The flow chart
for Pre-Processing stage is shown in Fig-1.
Fig-1: Steps in Pre-Processing Stage
The first step in pre-processing stage is extraction of frames
contained in the captured video. The resultant frames will be
RGB images. Working directly on a RGB image is
computationally inefficient because it involves large number
of working levels. To reduce the number of working levels we
convert these RGB images into grayscale images. This task is
accomplished using (1).
( , ) 0.229*R 0.587*G 0.114*BGray i j    (1)
Here Gray denotes the resultant output grayscale image while,
R, G, and B indicate the red, green and blue components of the
original RGB frame respectively [10]. To enable detection of
moving objects from complex background as well as dim light
environment we employ gray level stretching to highlight the
foreground pixels [10]. The resultant images from first stage
are shown in Fig-2, Fig-3 and Fig-4 respectively.
Fig-2: RGB Frame
Fig-3: Grayscale Frame
Captured
Video
Extraction of Frames
Contrast Enhancement
Further
Processing

__________________________________________________________________________________________
Fig-4: Result of Contrast Stretching
3.2 Exact Methodology
The flowchart for the proposed method is shown in Fig-5. To
eliminate the stationary background we consider two
consecutive frames. They are denoted as KF and 1KF 
respectively.
Fig-5: Flow Chart for Proposed Method
3.2.1 Edge Detection
In the next step we detect the edges present in each of these
frames. Three different operators have been analyzed for
detecting the edges present in frames. These are Prewitt’s
operator, Sobel’s operator and Canny’s operator [10]. The
result of edge detection using each of these three operators is
shown in Fig-6. As visible using Prewitt’s operator we may
miss the edges that represent the moving object, while using
Canny’s operator results in detecting extra edges apart from
the main object and will make our further processing task
complex. Hence in this approach we have considered Sobel’s
operator for detecting edges.
START
The (K-1)th the
Frame, FK-1
The Kth the
Frame, FK
Edge Detection using
Sobel’s operator to get
Edge Image EK-1
Edge Detection using
Sobel’s operator to get
Edge Image EK
Edge Difference Image
(ED)
Edge Density Based Region
Growing using a Window size
of 90×90
Component Labelling
and Connected
Component Analysis
Removal of components
which are too small
END

__________________________________________________________________________________________
(a) (a) Output of Prewitt’s operator (b) (b) Output of Sobel’s operator (c) Output of Canny’s Operator
Fig-6: Result of Edge Detection using different operators
The result of applying edge detection on frames FK and FK-1 is
shown in Fig-7 and Fig-8 respectively. After Edge detection,
the next step is to obtain the difference image to eliminate the
edges of stationary objects. This is accomplished using (2).
( 1) ( )K Sobel K SobelED E E  (2)
3.2.2 Edge Density Based Region Growing
After this the next step is to apply Edge Density Based Region
Growing Algorithm [11]. This is done to connect the regions
containing the moving object which has been highlighted by
computing edge difference. To implement this we count
number of edge pixels over a pre-defined window. This
number is then normalized by dividing it with total no of
pixels in the window. The resultant value is compared with a
threshold ( ). If the value is greater than the pre-defined
threshold, all the pixels in the window are assigned a value of
1 else they are set to 0.
3.2.3 Connected Component Labeling
After connecting the regions containing the moving object the
next step is to assign labels to these regions so that we can
extract them one by one. The labeling is done based on 8-
connectivity. Next we search for smallest rectangle which
accommodates each of these regions as shown in Fig-11.
Fig-7: Edge image (EK)
Fig-8: Edge Image (EK-1)

__________________________________________________________________________________________
Fig-9: Edge Difference Image
Fig-10 shows the result of applying Edge Density Based
Region Growing Algorithm on Edge Difference Image.
As we can observe, Fig-11 contains extra regions with
dimensions much smaller than that of the detected moving
object. To eliminate these regions we apply a series of
morphological opening operations. The resultant modified
image is shown in Fig-12.
Fig-10: Result of applying EDBRG
Fig-11: Connected Component Labeling
Fig-12: Modified Image after Morphological Opening
The region visible with white mark in Fig-12 is the region
which contains the moving object. The exact moving object is
extracted by retrieving corresponding pixels values from
original image as shown in Fig-13.
Fig-13: Detected Moving Object

__________________________________________________________________________________________
4. RESULTS AND DISCUSSIONS
The proposed method is implemented on Matlab 7.8.0,
R2009A, on a PC having Intel Core-i3 Third generation CPU
with 4-GB RAM capacity running on Widows-8 Operating
System. The size of the frames considered for processing is
480×640. Fig-14 shows additional simulation results. Two
different videos have been considered: 1) A women moving in
her office 2) A cylinder rolling on a flat platform. The total no
of frames analyzed in first video is 179 while that in second
video is 133. The frame pairs considered for simulation are
(15, 17), (35, 37), (119,121) respectively from first video and
(50, 52), (75, 77) and (90, 92) respectively from second video.
5. CONCLUSIONS
In this paper a unique method for detecting moving object
based on edge detection, frame difference and edge density
based region growing is presented. Use of contrast stretching
in pre-processing before edge detection makes the method
robust enough for detecting the moving objects in complex
backgrounds. The main advantage of the method is that it
takes into account the edges present in the frames and hence
operates directly on binary images. This reduces the
computation time as in a binary image we have to work with
only two levels. Further improvement can made by carrying
out experiments with different window sizes and thresholds
while applying the edge density based region growing
algorithm. As a part of future work we look forward to
integrate object tracking with the proposed algorithm.
FK-1 FK Detected Moving Object
Fig 14 (a): Additional Simulation Results at different positions; Human Being

__________________________________________________________________________________________
Fig-14 (b): Additional Simulation Results at different positions; Moving Cylinder
REFERENCES
[1]. Kim Changick, “Fast and Automatic Video Object
Segmentation and Tracking for Content-Based Applications”,
IEEE Transactions on Circuits and Systems for Video
Technology, Vol. 9, pp. 1945-1062, July-1999.
[2]. K. Wu, T. Xu, H.Zhang, J. Song, “Overview of Video-
Based Vehicle Detection Technologies”, Sixth International
Conference on Computer Science and Education, pp. 821-825,
August-2011.
[3]. R. Collins, A. Lipton, T. Kanade, “Introduction to Special
Section on Video Surveillance”, In the Proceedings of IEEE
Transactions on Pattern Analysis and Machine Intelligence,
pp. 745-746, 2000.
[4]. L. Maddalena, A. Petrosino, “A Self-Organizing
Approach to Background Subtraction for Visual Surveillance
Applications”, IEEE Transactions on Image Processing, Vol.
17, No. 7, pp. 1168-1177, 2008.
[5]. Zhu Minghan, and Luo Dayong, “Moving Objects
Detection and Tracking Based on Two Consecutive Frames
Subtraction Background Model”. Computer Measurement and
Control, 2006, pp. 1004-1006.
[6]. B. Horn, and B. Schunck, “Determining optical flow”.
Artificial Intelligence, 1981, vol. 17, pp.185-203.
[7]. Taskeed Jabid, Tahseen Mohammad, Tanveer Ahsan, M.
Abdullah-Al-Wadud, Oksam Chae, “An Edge-Texture based
Moving Object Detection for Video Content Based
Application”, In the Proceedings of 14th
International
Conference on Computer and Information Technology (ICCIT
2011), December-2011.
[8]. M.Sankari and C. Meena ,“Adaptive Background
Estimation and object detection applying in Automated visual
surveillance” (IJCSIS) International Journal of Computer
Science and Information Security, Vol. 8, No. 4, July 2010.
[9]. H.S. Mohana, Aswatha Kumar and Shiva kumar Malnad
College of Engineering, Hassan, Karnataka,” Vehicle
Counting And Classification Using Kalman Filter And Pixel
Scanner Technique And Its Verification With Optical Flow
Estimation”, Global Journal of Computer Science and
Technology 46 Vol. 10 Issue 8 Ver. 1.0 September 2010.
[10]. R.C Gonzalez, R.E Woods, Digital Image Processing, 3rd
Edition, Prentice Hall, 2007.
[11]. Jianbin Jiao, QixiangYe, Qingming Huang, “A
configurable method for multi-style license plate recognition”,
Pattern Recognition 42, 2009, pp: 358-369.

Applying edge density based region growing with frame difference for detecting moving objects in video surveillance systems

Recommended

More Related Content

What's hot (20)

Viewers also liked (20)

Similar to Applying edge density based region growing with frame difference for detecting moving objects in video surveillance systems (20)

More from eSAT Publishing House (20)

Recently uploaded (20)

Applying edge density based region growing with frame difference for detecting moving objects in video surveillance systems