Sei sulla pagina 1di 27

VIDEO MONITORING USING

SPATIO-TEMPORAL MODEL AND


ALERT SYSTEM FOR SECURITY
(DOMAIN: IMAGE PROCESSING)
Batch No :

A4

211411104002
211411104008
211411104013

Abinaya.R
Anitha.V
ArulMary.X
Guided By
Dr.S.Malathi M.E.,PhD.,
Professor

OBJECTIVE

To address the security issues in todays environment, an effective


background subtraction method using GAUSSIAN MIXTURE MODEL is
proposed.

The background spanning over both space and time gets updated
automatically.

The method involves motion detection and online processing in which it


automatically updates the background by capturing the dynamic scenes.

An alert message is sent to the users mobile once the object is detected.

EXISTING PROBLEM
The existing method uses Background Subtraction technique using threshold
approach with Mathematical Morphology for dynamic object detection.
Once the object detection was done the user is alerted via a SMS(SHORT
MESSAGE SERVICE).

DRAWBACKS:
The processing time per frame of the video is as low as possible.
Since it involves n number of frames and therefore the object detection is
time consuming.
The system works only in batch mode and hence it is not suitable for real-time
scenario or realtime object detection.

PROPOSED WORK
The Main aim of this method is to detect the motion in the surveillance
area. The surveillance camera captures the images and then we extract a
sequence of regular video bricks.
The motion detection is done using GAUSSIAN MIXTURE MODEL based on
foreground detection and kalman filter for tracking the detected object.

where wk are the weights, and N (x; k, k) is the normal density of mean k
and covariance matrix k = kI; I denotes the identity matrix.
Once the object is detected ,an alert message is sent to the users mobile
with notification.

LITRETURE SURVEY
Title and year

Author

Problem
statement

Method

Performance

Limitations

Layered
graph
matching
with
composite cluster
sampling.(2010)

Liang
et al.,

Lin.

To
find
an
unknown number
of corresponding
graph structures
in two images.

Sampling
the
positive
and
negative
edges
and assigning new
colors using MCMC
(Markov
chain
Monte
Carlo)
approach.

Automatic
learning
and
recognising
object categories
and their parts.

It does not deal


with
a
substantial
data set

Bayesian
robust
principal
component
analysis.(2011)

Xinghao
Ding.
et
al.,

To extract lowdimensional
information from
the
highdimensional data

The
Bayesian
framework
and
exploitation
of
additional structure
in the matrix.

Robust to densely
distributed noise.

Background on
low
dimensional
manifold.

LITRETURE SURVEY
Title and year

Author

Problem
statement

Method

Performance

Limitations

Integrating
spatio-temporal
context
with
Multiview
representation
for
object
recognition
in
visual
surveillance.
(2011)

Xiaobai
Liu. et al.,

Rapid
recognition of
dynamic
objects by
exploring
various
contextual
information.

Multiview
object
representation;
unified
spatiotemporal
context
model;
samplingbased
inference
procedure.

Can process 7 to
10 frames/s

Scene-specific
and thus not
flexible enough
for
moveable
system.

Integrating
graph
partitioning and
matching
for
trajectory
analysis
in
video
surveillance.
(2012)

Liang Lin.
Et al.,

Track moving
objects
in
long-range
against
occlusion,
interruption
and
background
clutter.

Foreground map by
a
joint
task
of
spatial
graph
partitioning
and
temporal
graph
matching.

Can process 3 to 5
frames/s

Needs
improvement
to study the
parallel
implementatio
n,
for
the
MCMC-based
inference.

LITRETURE SURVEY
Title and year

Author

Problem
statement

Method

Performance

Limitations

Contextualized
trajectory
parsing with
Spatio-temporal
graph(2013)

Xiaobai
Liu. et al.,

To
solve
spatial
segmentation,
temporal tracking,
and
Object
categorization.

Explains
the
novel
representation
of
spatiotemporal graph
with
nodes
expressing the
motion
primitives.

High-quality
Tracking

To extend STgraph for other


image
tasks,
image
segmentation,
and
object
detection.

Complex
Background
subtraction
by
Spatio-Temporal
model(2014)

Liang Lin.
et al.,

Background
subtraction is still
an open problem in
the
context
of
complex
scenarios.for
eg,
dynamic
backgrounds
and
illumination
variations.

Background
Subtraction
method
by
learning
and
maintaining an
array of dynamic
texture models
with in a SpatioTemporal
context.

Process
15-20
frames/second in
the
resolution
352x288(pixels)

Pixel
comparison with
color images is
not
accurate
and may lead to
misassumption
of objects with
similar color.

DESIGN-ARCHITECTURE
DIAGRAM

FIG 1. ARCHITECTURE DIAGRAM OF


THE PROPOSED METHOD

LEVEL-0 DATA FLOW


DIAGRAM

FIG 2. LEVEL-0 DATA FLOW DIAGRAM

LEVEL-1 DATA FLOW


DIAGRAM

FIG 3. LEVEL-1 DATA FLOW DIAGRAM

STATE TRANSITION
DIAGRAM

FIG 4. STATE TRANSITION DIAGRAM

MODULE DESCRIPTION
MODULE 1: OBJECT DETECTION
This module presents object detection using Gaussian Mixture model based
foreground detection and kalman filter for tracking the object.

MODULE 2: ALERT SYSTEM


This module presents an alert message sent to the user once the object is
detected along with the notifications.

IMPLEMENTATION OF
MODULES
MODULE 1: OBJECT DETECTION
The input video frame is sent for processing to detect and track the object.
The object is detected using Gaussian Mixture Model (GMM).
A GMM is a parametric probability density function represented as a
weighted sum of Gaussian component densities.
p( x) ~ N ( , 2 ) :
1 ( x )2

1
p( x)
e 2
2

The GMM uses parameters such as mean() and co-variance() to calculate


threshold value.
With the help of threshold value,comparison is made and object is detected.

ALGORITHM-1
Name:
Algorithm for Object Detection;
Input:
Video Frames; Output: Detected Object;
Step1:
InitialiseBG={},numFrames=1,ResetmaxFrameflag,Vi
maxNumFrames};
Step2:
Data : input video x(t)
Step3:
Calculate i, i;
Step4:
while New Data x(t) do
for ic{1.numFrames} do
if(x(t),iframe matches mode i+1) then
if i C BG then
x(t) is foreground;
update background;
ii+1;
update i, i;
if(numFrames<maxNumFrames) then
numFramesnumFrames+1;
initialize newframe j;
Step5:

Compare the threshold value,


if(x(t) is background) then
0x(t);
elseif (x(t) is foreground) then
1x(t);
else
reset maxFrameflag;

belongs

to{1

IMPLEMENTATION OF
MODULES
MODULE 2 : OBJECT TRACKING
The tracking of object is done using Kalman filter.
Kalman filter algorithm uses series of
containing noise and other inaccuracies.
This algorithm involves two steps,
(i) Prediction
(ii) Correction

measurements observed over time,

FORMULA-1
TIME UPDATE EQUATION (PREDICT)
Step 1.1: Project the state ahead
Step 1.2 : Project the error covariance ahead

MEASUREMENT UPDATE EQUATION


(CORRECT)
Step
2.1: Compute the kalman gain (K)

Step 2.2: update estimate with measurement Zk


Step 2.3: update the error covariance
k

Where, =update step;


=acceleration;
Z=update state;
H=predicted covariance;

=predicted step;
=predicted state;
=updated covariance;
Q=sensor noise;

IMPLEMENTATION OF
MODULES
MODULE 3 : ALERT SYSTEM
The alert is produced when an object is detected in the system.

The alert is given in the form of an SMS alert and mail containing the image

of the detected

object.

The premises are alerted by the system with the help of a buzzer alarm.

The buzzer alarm is triggered on detection of moving object by the system.

PSEUDOCODE-1
Name: Alert system
Input:

Detected object video frame;

Output: Notification is users Mobile


Step 1: if( object detected)
send alert message to users mobile;
buzzer alarm is activated;
Step 2: else
continue operation( module 1);
end

RESULTS AND DISCUSSIONS


S.No

Input

Filter
name

Elapsed
time
(sec)

Specificity
in %

1.

.avi file

Canny
filter

5.5

98.5

2.

.avi file

Mean filter

4.9

99.8

3.

.avi file

Weiner
filter

5.2

93.1

4.

.avi file

Prewit
filter

5.2

98.6

5.

.avi file

Robert
filter

5.3

99.1

6.

.avi file

Kalman
filter

4.3

100

TABLE 1. EXPERIMENTAL EVALUATION OF


VARIOUS FILTERS

RESULTS AND DISCUSSIONS


Object
Human walking in
garage
Human walking in
beach
Human in high
lighting change
Human walking up
hill
Human in Super
Market

Background
Static

Sensitivity in %
100

Dynamic

93.6105

Dynamic

83.8494

Dynamic

80.4060

Static

52.3658

TABLE 2. EXPERIMENTAL EVALUATION OF MOTION DETECTION


Sensitivity in %
150
100
50
0

Sensitivity

FIG.5 EVALUATION OF MOTION DETECTION

SCREENSHOTS

FIG 6. OBJECT DETECTION AND TRACKING FOR


HUMAN WALKING UP HILL

SCREENSHOTS

FIG 7. OBJECT DETECTION AND TRACKING FOR HUMAN IN HIGH LIGHTING CHA

SCREENSHOTS

FIG 8. OBJECT DETECTION AND TRACKING FOR HUMAN WALKING IN GARAGE

SCREENSHOTS

FIG 9. EVALUATION PARAMETERS

FIG 10.ALERT WITH AN IMAGE

CONCLUSION

The proposed method of motion detection features high accuracy of


moving object detection and quick notifications to the user on the
intrusion.

The advantages of this system are it is less sensitive to noise.

The background for the comparison does not use reference background but
gets updated at certain intervals based on the previous incoming frame.

In future, the focus much less time for object detection and alerts directly
given to the police will be enhanced.

REFERENCES
1.

Liang Lin, Yuanlu Xu, Xiaodan Liang and Jianhuang Lai, Complex Background
Subtraction By Pursuing Dynamic Spatio-Temporal Models IEEE Transactions on
Image Processing. Vol. 23, Issue: 7, 2014

2.

Tang Chunrai;Lui Dandan;Yanru.S; The AUV Movement Target detection based on


LIP Model at ststic background2013 Fifth International Conference on Measuring
Technology And Mechatronics Association(ICMTMA),pp.1139-1141,2013.

3.

Zhu,Q;Zhan Song ; Xie, Y.;Wang,L. A Novel Recursive Bayseian Learning-Based


Method for the Efficient and Accurate Segnmentation of Video with Dynamic
Background IEEE Transactions on Image Processing ,Volume:21,Issue:9, pp.38653876,2012.

4.

X.Liu,

L.Lin,

S.Yan,

Integrating

spatio-temporal

context

with

multiview

representation for object recognition in visual surveillance,IEEE Trans.Circuits


Syst.Video Technol.,vol. 21, no. 4, pp. 393-407, Apr. 2011.
5.

Z. Zhou, A. Wagner, H. Mobahi, J. Wright, and Y. Ma, Face Recognition with


Contiguous Occlusion Using Markov Random Fields, Proc. IEEE Intl Conf.
Computer Vision, 2010.

PUBLICATIONS
INTERNATIONAL JOURNAL

Anitha.V, Abinaya.R, Dr. S.Malathi, Video Monitoring using Spatio-Temporal


Model and Alert System for Security, International Journal of Intelligent
Information Technologies (IJIIT),vol.11,2015

INTERNATIONAL CONFERENCE

Anitha.V, Abinaya.R, Dr. S.Malathi, Video Monitoring using Spatio-Temporal


Model and Alert System for Security International Conference on
Intelligent Information Technologies[ICIIT]-2014,Anna University, Dec 2014.

Potrebbero piacerti anche