1

Road Crosser Warning Signal System Using

Digital Video Image Processing

Supriya.T 1 Selvakumar.R.K2 and patra.P.S.K3

1
Department of Computer Science and Engineering, 2Professor&Head, Department of Information Technology,
3Professor&Head, Department of computer science & engineering,

Agni College of Technology,Anna University,Chennai,Tamilnadu ,

[email protected], [email protected], [email protected]

loitering or moving into an unauthorized area are

automatically highlighted and forwarded to security
Abstract persons. In moving, cameras can capture and recognize
Object detection is the fundamental step in video analysis in
license plate numbers for enforcement and toll collection
many applications. Existing motion based methods are limited
to detect the moving object when coping with dynamic purposes. In trade, video analytics can count the number
background and non rigid foreground motion. In this we of people waiting in line.
optimize the object detection and background learning into a
single process. This work is on people tracking followed by their 2.INTRUSION DETECTION AND
face recognition. Face recognition presents a challenging UNATTENDED BAGGAGE DETECTION
problem in the field of image analysis and computer vision
especially when it is dealt with video sequences. Also in this
project, we propose a system for identify the road crosser either
human or animal through videos, based on shape analysis Intrusion detection provides surveillance watch
method. This project helps to prevent the accidents happening in automation that automatically detects prohibited intrusion
India. scenarios and can be used on stationary cameras. There
.
are several modes for detection: Regional Entrance,
Tripwire and Fence Trespassing.A Provides an alarm
1.INTRODUCTION when an item (package, debris, baggage, etc.) is deposited
or appears in a controlled area. In addition, recognition
Video Analytics is enabling a rapidly growing number of can be configured to ignore items that are attended by a
embedded video products such as smart cameras and nearby person. Helps reduce the need for patchy roving
intelligent Digital Video Recorders (DVRs) with patrols and provides rapid detection and pre-alarm
automated capabilities that just a few years ago would recording so that abandon luggage owners can be located
have required human observing. Generally, video more rapidly and also provides detection for disguised
analytics is the pulling out of meaningful and relevant objects that may be overlooked by passing patrols or
information from digital video. As per conflicting to video regarded as familiar. With rapid response and other
compression, which attempts to achievement information on how the baggage arrived, responders have
the severance in digital video for the purpose of more time and benefit from greater situational awareness
decreasing size, analytics is afraid with understanding the for scene safety and decision-making.
contented of video. Video Analytics forms research
in computer vision, pattern analysis and machine
intelligence, and spans several production sections 3. SEGMENTATION
including surveillance, trade and transportation. It is also
called intelligent video. Segmentation is the process of identifying components of
the image. Segmentation includes operations such as
Video analytics is used for automatic observation. Smart boundary detection, connected component labeling,
cameras with analytics continuously analyze video and
thresholding etc. Edge detection finds out boundaries in
can detect the presence of people and vehicles and the image. Any difference operator can be used for edge
interpret their activities. Doubtful activities such as
 2

detection. Thresholding is the process of reducing the extraction. The important two methods are the use of
grey levels in the image. Many algorithms happen for difference image and kalman filtering.
thresholding. The aim of image segmentation algorithms .
is to partition the image into perceptually related areas.
Every segmentation algorithm addresses two problems,
4.1 Use Of Difference Images
the conditions for a good partition and the method for
achieving efficient partitioning . In this section, we will Some of the methods for foreground extraction are Use of
discuss recent segmentation techniques that are relevant difference images In this method we use subtraction of
to object tracking. images in order to find objects that are moving and those
that are not. The result of the detraction is viewed as
another grey image called difference image. Three types
3.1 Mean Shift Clustering of difference images are defined

For the image segmentation problem, the mean-shift

• Absolute accumulative difference image is given
method to find clusters in the joint spatial and color space,
[l , u, v, x, y], where [l , u, v] represents the color and [x, by
y] represents the spatial location. Given an image, the
algorithm is modified with a large number of f(x,y) = f(x,y) +1 ……………….if |g(x,y,ti+1) -
hypothesized cluster centers randomly chosen from the
data. Then, each cluster center is moved to the mean of g(x,y,ti)| > T
the data lying inside the multidimensional ellipsoid
centered on the cluster focal point. The vector defined by • Positive accumulative difference image is given by
the old and the new cluster centers is called the mean-
shift vector. The mean-shift vector is calculated iteratively
until the cluster centers do not change their points. Note f(x,y) = f(x,y) +1 ……………….if g(x,y,ti+1) -
that during the mean-shift iterations, some clusters may g(x,y,ti) > T
get merged. The segmentation using the mean-shift
approach generated using the source code available at
Mean Shift Segments. Mean-shift clustering is scalable to • Negative accumulative difference image is given by
various other applications such as edge detection, image
regularization , and tracking .Mean-shift based f(x,y) = f(x,y) +1 ……………….if g(x,y,ti) -
segmentation requires fine tuning of various parameters to
g(x,y,ti+1) > T
obtain better segmentation, for instance , selection of
the color and spatial kernel bandwidths , and the
threshold for the minimum size of the region considerably
effect the resulting segmentation. 4.2 Kalman Filtering
3.2 Image Segmentation Using Graph-Cuts
‘kalman filter’ for calculating the image at ti+1 based on
some noise model. The difference between expected and
Image segmentation can also be formulated as a graph actual attentions is thresholded to classify the image pixel
partitioning problem, where the vertices (pixels), of a as foreground or background. One benefit of this method
graph (image), centroid, are partitioned into N disjoint is it considers effect of noise, which is very important
sub graphs , by pruning the weighted edges of the graph. feature in real world applications.
The total weight of the pruned edges between two sub Kalman filtering is composed of two steps, prediction and
graphs is called a cut. The weight is typically computed correction. The prediction step uses the state model to
by the color, brightness, or texture similarity between the predict the new state of the variables. Equally, the
nodes. correction step uses the current observations to update the
object’s state. The Kalman filter has been extensively used
4 . FOREGROUND EXTRACTION in the vision community for tracking.

As the name suggests this is the process of separating the 5. BACKGROUND EXTRACTION
foreground and background of the image. Now it is
supposed that foreground contains the objects of concern.
There are different methods used for foreground Another method that can be used in object tracking is
Background learning. This method can be used when
 3

fixed cameras are used for video capturing. In this changes, and finally in the real world there is always
method, an initial exercise step is carried out before noise. Furthermore, from semantic point view, in scenes
installing the system. In the training step the system containing motion there are occlusions among the objects,
constantly records the background in order to ‘learn’ it. as well as disappearing of objects and appearing of new
Once the training is complete the system has complete (or ones. Despite the problems of pixel by pixel
almost complete) information about the background. And correspondence, it is fast to compute and is used
yet this step is slightly long, it has a very important extensively for finding matching regions. Some of the
benefit. Once we know the background, extracting the most often used matching criteria based on pixel
foreground is matter of simple image subtraction. differencing are mean absolute distance (MAD), mean
squared distance(MSD).
Object detection can be achieved by building a
representation of the scene called the background model 6.3 Exploiting The Domain Knowledge
and then finding deviations from the model for each
received frame. Any significant change in an image
region from the background model signifies a moving S. Russell and J.Weber discussed a different
object. The pixels founding the regions undergoing approach in which domain knowledge is exploited to
change are marked for further processing. Generally, a simplify object tracking . As the objects under
connected component algorithm is applied to obtain consideration are vehicles, image path can be
connected regions corresponding to the objects. This approximated by affine transformations. Since motion is
process is denoted to as the back ground subtraction. constrained to the road plane and since possible rotation
components along the normal of the plane are small the
degrees of freedom can be reduced to the extent that we
6. OBJECT TRACKING obtain a velocity equation of only a scale parameter s and
a displacement vector u(x):
The next step is to extract useful features from the
arrangement of frames. Depending on the algorithm, U(X) = S ( X – Xm ) + U0
definition of ‘feature’ can differ. Numerous approaches for
object tracking have been proposed.
Here, s is the scaling factor. When s = 0 there is no
6.1 Feature Based Method scaling. When s>0, the motion is towards the camera and
when s< 0, it is away from the camera. xm denotes the
Selecting the right features plays a critical role in the centre of the moving image region and u0 denotes its
tracking. In general, the most desirable property of a displacement between two consecutive frames. Third
visual feature is its uniqueness so that the objects can be Kalman filter is used to estimate motion parameters.
easily distinguished in the feature space. Feature
selection is closely related to the object representation.
For an example, color is used as a feature for histogram-
based appearance representations, while for contour-based 6.4 Compressed Domain Object Trackin.
representation, object edges are usually used as the
features. This tracking method uses compressed domain MPEG
video as the source. If input is compressed domain data
6.2 The Block Matching Algorithm then the system needs to spend extra time to first
uncompress the data before processing them. This
The block matching algorithm is a standard technique for further reduces system performance. Thus having
encoding motion in video sequences . It aims at detecting uncompressed input is desirable in real time system.
the motion between two images in a block-wise sense. The
blocks are usually defined by dividing the image frame
into non-overlapping square parts. Each block from the 7. Conclusions
current frame is matched into a block in the destination
frame by shifting the current block over a predeframe. At
each shift, the sum of the distances between the gray From the discussion, it can be seen that object tracking
values of the two blocks is computed the sum of the has many useful applications in the robotics and
distances between the gray values of the two blocks is computer vision fields. Several researchers have explored
computed. The shift which gives the smallest total and implemented different approaches for tracking. The
distance is considered the best match. In the ideal case, success of a particular approach depends largely on the
two matching blocks have their corresponding pixels problem domain. In other words, a method that is
exactly equal. This is rarely true because moving objects successful in robot navigation may not be equally
change their shape in respect to the observer's point of successful in automated surveillance or in traffic
view, the light reflected from objects' surface also monitoring. Further there exists a cost/performance
 4

trade off. For real time applications we may need a fast

high performance system on the other hand offline
applications we may use a relatively cheap (and slower in
performance). It can also be seen from the diverse nature
of the techniques used that the field has a lot of room for
improvement.

8.References

[1] R. Gonzalez and R. Woods – Digital image processing

– second edition – Prentice Hall.

[2] L. G. Shapiro and R. M. Haralick – Computer and

robot vision – volume 1 and 2 – 1 st edition – Prentice
Hall.

[3] Y. Wang, J. Doherty and R. Van Dyck – Moving object

tracking in video - Proc. Conference on Information
Sciences and Systems, Princeton, NJ, March 2000.

[4] C. Ridder, O. Munkelt, and H. Kirchner - Adaptive

Background Estimation and Foreground Detection
using Kalman-Filtering - Proceedings of International
Conference on recent Advances in Mechatronics
(ICRAM), pp. 193-199, 1995.
[5] A. Gyaourova, C. Kamath, S. and C. Cheung - Block
matching object tracking - LLNL Technical report,
October 2003

[6] Y. Rosenberg and M. Werman – Real-Time Object

Tracking from a Moving Video Camera: A software
approach on PC - Applications of Computer Vision,
1998. WACV '98. Proceedings.

[7] A. Turolla, L. Marchesotti and C.S. Regazzoni –

Multicamera object tracking in video surveillance
applications.

[8] Çiˇgdem Eroˇglu Erdem and Bülent San - Video

Object Tracking With Feedback of Performance
Measures - IEEE Transactions on circuits and systems
for video technology, vol. 13, no. 4, April 2003

[9] T. Huang, D. Koller, J. Malik, G. Ogasawara, B. Rao,

S. Russell and J. Weber - Automatic Symbolic Traffic
Scene Analysis using Belief Networks - Third
European Conference on Computer Vision, Stockholm,
Sweden, May 1994

[10] G. Welch and G. Bishop - An Introduction to the

Kalman Filter – May 2003

[11] R. Achanta, M. Kankanhalli, P. Mulhem - Compressed

domain object tracking for automatic indexing of
objects in MPEG home video - Proc. IEEE
International conference in Multimedia and Expo
(ICME 2002), Lausanne, August 2002.