Human Detection in Video Surveillance System
Content
International Journal on Recent and Innovation Trends in Computing and Communication
Volume: 3 Issue: 4
ISSN: 2321-8169
2171 – 2173
_______________________________________________________________________________________________
Human Detection in Video Surveillance System
1
Ajinkya Sapkal, 2Sushil Nemade,3 Nikhil Mohadikar, 4Pallavi Gosavi
1,2,3,4
Department of Computer Engineering Sinhgad College of Engineering, Pune, India
Email: [email protected],[email protected],[email protected],[email protected]
Abstract— Object detection is a crucial part in today’s video surveillance systems. Many methods have evolved over the years that include
Background Subtraction at the pinnacle. Background subtraction is a technique in which the video is segmented in multiple frames. A base frame
called as “Background” is used to subtract another frame from it to detect “Foreground”. Motion–based and shape-based algorithms boost the
Background subtraction method.
The multiple objects detection technique used in surveillance system uses Support Vector Machine (SVM) to detect and classify the
different objects. In this project, study proposes a novel object detection and its classification using Support Vector Machine (SVM) which is
used to differentiate objects according to the set of points on the objects. The algorithm then aims at the classification of these key-points, namely
at discriminating between the points which belongs to objects and all the others, by means of a Support Vector Machine (SVM) classifier. At the
end of the procedure, the objects present inside the scene are identified by analyzing at the key-points previously classified as specific object
points. It begins with a feature extraction process from which a set of consistent key-points is identified. Being able to identify specific objects or
a particular class of objects in an image can provide several advantages and can open the door to the development of various interesting
applications.
Index Terms— Human detection, Image Segmentation, Feature extraction, Support vector machine (SVM) classification
__________________________________________________*****_________________________________________________
III.
I.
Object detection has evolved over the years. There
have been many advances in the technology of object
detection and classification. Object detection and
classification is a very key part in today’s surveillance
systems. Technology can be classified into mainly three
generations namely 1GSS, 2GSS, 3GSS.
1GSS: This generation used analog systems for Image
processing and the retrieval of data was a tedious job
2GSS: This generation used both analog and digital
systems which improved the speed of object detection
3GSS: This generation used full-fledged digital systems for
image processing and object classification. This proved the
retrieval of the data to be fast and a central repository reduced
the cost of the systems. The proposed system targets human
detection and it is database management in a video
surveillance system for security purposes and marketing
strategies. The rest of this paper is organized as follows.
Sections II and III introduces the motivation and main idea
respectively. Section IV and V are preprocessing and feature
extraction from a particular frame respectively. Section VI
consist of the classification module.
II.
MAIN IDEA
INTRODUCTION TO VSS
MOTIVATION
This work is motivated by the ever increasing
application of image processing and high capacity storage
devices used in places such as military services, traffic
surveillance etc.
Also, the technology provides a useful application in
modern vehicles such as Unmanned Aerial Vehicles (UAV)
There is increasing demand for data analysis in the
field of security systems, banking environment, forensics etc.
There are three main jobs in this system. The jobs are
preprocessing, extraction of features and feature
matching/classification using SVM (support vector machine).
Figure 1. Block Diagram
Aim is to extract a complete human body region from
a video. The difficulties are in determining a human region in
frames and to find the correct human regions.
First frame is taken as initial background. Input
frames are subtracted from the background to generate a
coarse mask of foreground regions. Then connected
components searching and operations are applied on the mask
to obtain refined foreground regions. Subsequently, we collect
sample points on the contour of a foreground region and from
which shape feature vectors are abstracted.
An SVM classifier is used for identifying whether the
input vectors belong to human or nonhuman based on the
model previously trained by positive and negative samples.
Every independent foreground region is classified and regions
identified as humans are recorded for updating background.
2171
IJRITCC | April 2015, Available @ http://www.ijritcc.org
_______________________________________________________________________________________
International Journal on Recent and Innovation Trends in Computing and Communication
Volume: 3 Issue: 4
ISSN: 2321-8169
2171 – 2173
_______________________________________________________________________________________________
IV.
PREPROCESSING
The main of preprocessing is an improvement of the
image data that suppresses unwanted distortion and enhances
some image features important for further processing. Input
video is converted into frames and these frames are converted
into grey images.
Preprocessing consist of converting video into frames
and background subtraction. Background subtraction is based
on frame differencing. Background subtraction attempts to
detect moving objects from the difference between the current
frame and the reference frame.
width ratio) is a reliable source as aspect ratio of any human
being is fairly greater than 1. Fig 3.(a) background subtraction
result is used for further feature extraction. Fig 3.(b) shows a
contour around the foreground mask in a form of green
rectangle. This rectangle is used for calculating height and
width of the extracted object. Aspect ratio is calculated frame
by frame as,
Aspect ratio = Height/Width
(a)
Fig 4. Face Detection
(b)
Fig 2. Preprocessing
Fig 2 shows actual preprocessing of a video. Fig 2.
(a) is the base frame and (b) is the current frame taken for
background subtraction. One can see the actual result of
background subtraction in Fig 2. (c). Thus, the resulting image
will contain information about how much changes occurred
between the two frames.
V.
FEATURE EXTRACTION
The purpose of image segmentation is to separate
foreground regions from background area in order to detect
any moving objects. Main concern is the foreground region for
further processing after background subtraction.
Fig 3.Feature Extracted
Correct features for human detection are necessary
for accurate results. Feature such as aspect ratio (height to
Living things and non-living things are generally
differentiated on the basis of facial feature. Often, non-living
objects do not have faces. This feature is used along with
Aspect ratio so as to eliminate more objects other than
humans. Fig 4.(a) shows two persons and their faces detected.
Pink-colored ellipse is drawn around person’s face. Fig 4.(b)
shows three persons and their faces detected on the basis of
frontal-face. For face detection, two eyes are required to be
seen in the frame.
VI.
CLASSIFICATION
In recent years, SVM raised so much importance in
the field of image processing because of its accuracy. As our
matter is to classify objects into human or non-human is a two
class problem SVM is the best known in finding solution to
such problem.
SVM is the binary classifier. As shown in Fig 5 it
classifies detected objects into two classes i.e. human and nonhuman. Hyperplane is constructed to differentiate between two
classes.
Fig 5 SVM Classifier
The line differentiates green and blue region is
hyperplane. The hyperplane is equidistant from two closest
2172
IJRITCC | April 2015, Available @ http://www.ijritcc.org
_______________________________________________________________________________________
International Journal on Recent and Innovation Trends in Computing and Communication
Volume: 3 Issue: 4
ISSN: 2321-8169
2171 – 2173
_______________________________________________________________________________________________
points from both the classes. Blue portion consist of nonhuman objects and green portion is for humans.
VII.
EXPERIMENTS RESULTS
We have tested the proposed human detection
technique on a number of videos. These videos are of various
real time scenarios.
Single human videos: Only single person appears. Videos have
different backgrounds.
Multiple human videos without occlusion: These videos have
multiple humans but they move in such a way that occlusion
doesn't occur. The study shown the human detection result
with video consisting of human and non-human objects at a
resolution of 720*1080, a frame rate of 22fps and for effective
calculation purpose we skipped 10 frames per second. The
video consisted of around 300-400 frames. One can observe
from Fig 5 that the proposed method classifies objects into two
classes from one considered frame accurately.
VIII.
SHIMOSAKODA, and Satoshi NAKAGAWA “Automated
Detection of Human for Visual Surveillance System” 1996 IEEE
Proceedings of ICPR '96
[9] Josef Grahn, “Using SVM for Efficient Detection of Human
Motion” Proceedings 2nd Joint IEEE International Workshop on
VS-PETS, Beijing, October 15-16, 2005
[10] Soumya Varma, Sreeraj M “Object Detection and Classification
in Surveillance
System” 2013 IEEE Recent Advances in
Intelligent Computational Systems (RAICS)
CONCLUSION
Nowadays various sectors are facing the security
problem hence we introduced this system so that human can
be detected in video surveillance system.
Input video is preprocessed i.e. converted into frames and
background subtraction is performed. Foreground objects are
obtained from background subtraction. Features are extracted
from these objects. These features are given as input to the
SVM classifier which classifies the object into Human or nonhuman objects
REFERENCES
[1] Manish Khare, Rajneesh Kumar Srivastava and Ashish Khare,
“Automatic
Multiple Human Detection and Tracking for
Visual Surveillance System”, IEEE/OSAIIAPR International
Conference on Informatics, Electronics & Vision
[2] Darin Brezeale and Diane J. Cook, Senior Member, IEEE
“Automatic Video Classification: A Survey of the Literature”,
IEEE TRANSACTIONS ON SYSTEMS, MAN, AND
CYBERNETICS, VOL. UNKNOWN, NO. UNKNOWN,
UNKNOWN 2007
[3] Ruofeng Tong, Di Xie, and Min Tang “Upper Body Human
Detection and Segmentation in Low Contrast Video” IEEE
TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR
VIDEO TECHNOLOGY, VOL. 23, NO. 9, SEPTEMBER 2013
[4] Hao Su1, Min Sun, Li Fei-Fei, Silvio Savarese “Learning a dense
multi-view representation for detection, viewpoint classification
and synthesis of object categories”
2009 IEEE 12th
International Conference on Computer Vision (ICCV) 978-14244-4419-9/09/$25.00 ©2009 IEEE
[5] Thomas MORANDUZZO and Farid MELGANI “A SIFT-SVM
METHOD FOR DETECTING CARS IN UAV IMAGES” 9781-4673-1159-5/12/$31.00 ©2012 IEEE
[6] Qiang Chen1, Zheng Song, Zhongyang Huang, Yang Hua and
Shuicheng Yan “Contextualizing Object Detection and
Classification” IEEE TRANSACTIONS ON PATTERN
ANALYSIS AND MACHINE INTELLIGENCE, VOL. XX,
NO. XX, 2013
[7] A. Ess1, B. Leibe, K. Schindler, L. van Gool “Moving Obstacle
Detection in Highly Dynamic Scenes” 2009 IEEE International
Conference on Robotics and Automation Kobe International
Conference Center Kobe, Japan, May 12-17, 2009
[8]
Yuji
KUNO,
Takahiro
WATANABE,
Yoshinori
2173
IJRITCC | April 2015, Available @ http://www.ijritcc.org
_______________________________________________________________________________________
Volume: 3 Issue: 4
ISSN: 2321-8169
2171 – 2173
_______________________________________________________________________________________________
Human Detection in Video Surveillance System
1
Ajinkya Sapkal, 2Sushil Nemade,3 Nikhil Mohadikar, 4Pallavi Gosavi
1,2,3,4
Department of Computer Engineering Sinhgad College of Engineering, Pune, India
Email: [email protected],[email protected],[email protected],[email protected]
Abstract— Object detection is a crucial part in today’s video surveillance systems. Many methods have evolved over the years that include
Background Subtraction at the pinnacle. Background subtraction is a technique in which the video is segmented in multiple frames. A base frame
called as “Background” is used to subtract another frame from it to detect “Foreground”. Motion–based and shape-based algorithms boost the
Background subtraction method.
The multiple objects detection technique used in surveillance system uses Support Vector Machine (SVM) to detect and classify the
different objects. In this project, study proposes a novel object detection and its classification using Support Vector Machine (SVM) which is
used to differentiate objects according to the set of points on the objects. The algorithm then aims at the classification of these key-points, namely
at discriminating between the points which belongs to objects and all the others, by means of a Support Vector Machine (SVM) classifier. At the
end of the procedure, the objects present inside the scene are identified by analyzing at the key-points previously classified as specific object
points. It begins with a feature extraction process from which a set of consistent key-points is identified. Being able to identify specific objects or
a particular class of objects in an image can provide several advantages and can open the door to the development of various interesting
applications.
Index Terms— Human detection, Image Segmentation, Feature extraction, Support vector machine (SVM) classification
__________________________________________________*****_________________________________________________
III.
I.
Object detection has evolved over the years. There
have been many advances in the technology of object
detection and classification. Object detection and
classification is a very key part in today’s surveillance
systems. Technology can be classified into mainly three
generations namely 1GSS, 2GSS, 3GSS.
1GSS: This generation used analog systems for Image
processing and the retrieval of data was a tedious job
2GSS: This generation used both analog and digital
systems which improved the speed of object detection
3GSS: This generation used full-fledged digital systems for
image processing and object classification. This proved the
retrieval of the data to be fast and a central repository reduced
the cost of the systems. The proposed system targets human
detection and it is database management in a video
surveillance system for security purposes and marketing
strategies. The rest of this paper is organized as follows.
Sections II and III introduces the motivation and main idea
respectively. Section IV and V are preprocessing and feature
extraction from a particular frame respectively. Section VI
consist of the classification module.
II.
MAIN IDEA
INTRODUCTION TO VSS
MOTIVATION
This work is motivated by the ever increasing
application of image processing and high capacity storage
devices used in places such as military services, traffic
surveillance etc.
Also, the technology provides a useful application in
modern vehicles such as Unmanned Aerial Vehicles (UAV)
There is increasing demand for data analysis in the
field of security systems, banking environment, forensics etc.
There are three main jobs in this system. The jobs are
preprocessing, extraction of features and feature
matching/classification using SVM (support vector machine).
Figure 1. Block Diagram
Aim is to extract a complete human body region from
a video. The difficulties are in determining a human region in
frames and to find the correct human regions.
First frame is taken as initial background. Input
frames are subtracted from the background to generate a
coarse mask of foreground regions. Then connected
components searching and operations are applied on the mask
to obtain refined foreground regions. Subsequently, we collect
sample points on the contour of a foreground region and from
which shape feature vectors are abstracted.
An SVM classifier is used for identifying whether the
input vectors belong to human or nonhuman based on the
model previously trained by positive and negative samples.
Every independent foreground region is classified and regions
identified as humans are recorded for updating background.
2171
IJRITCC | April 2015, Available @ http://www.ijritcc.org
_______________________________________________________________________________________
International Journal on Recent and Innovation Trends in Computing and Communication
Volume: 3 Issue: 4
ISSN: 2321-8169
2171 – 2173
_______________________________________________________________________________________________
IV.
PREPROCESSING
The main of preprocessing is an improvement of the
image data that suppresses unwanted distortion and enhances
some image features important for further processing. Input
video is converted into frames and these frames are converted
into grey images.
Preprocessing consist of converting video into frames
and background subtraction. Background subtraction is based
on frame differencing. Background subtraction attempts to
detect moving objects from the difference between the current
frame and the reference frame.
width ratio) is a reliable source as aspect ratio of any human
being is fairly greater than 1. Fig 3.(a) background subtraction
result is used for further feature extraction. Fig 3.(b) shows a
contour around the foreground mask in a form of green
rectangle. This rectangle is used for calculating height and
width of the extracted object. Aspect ratio is calculated frame
by frame as,
Aspect ratio = Height/Width
(a)
Fig 4. Face Detection
(b)
Fig 2. Preprocessing
Fig 2 shows actual preprocessing of a video. Fig 2.
(a) is the base frame and (b) is the current frame taken for
background subtraction. One can see the actual result of
background subtraction in Fig 2. (c). Thus, the resulting image
will contain information about how much changes occurred
between the two frames.
V.
FEATURE EXTRACTION
The purpose of image segmentation is to separate
foreground regions from background area in order to detect
any moving objects. Main concern is the foreground region for
further processing after background subtraction.
Fig 3.Feature Extracted
Correct features for human detection are necessary
for accurate results. Feature such as aspect ratio (height to
Living things and non-living things are generally
differentiated on the basis of facial feature. Often, non-living
objects do not have faces. This feature is used along with
Aspect ratio so as to eliminate more objects other than
humans. Fig 4.(a) shows two persons and their faces detected.
Pink-colored ellipse is drawn around person’s face. Fig 4.(b)
shows three persons and their faces detected on the basis of
frontal-face. For face detection, two eyes are required to be
seen in the frame.
VI.
CLASSIFICATION
In recent years, SVM raised so much importance in
the field of image processing because of its accuracy. As our
matter is to classify objects into human or non-human is a two
class problem SVM is the best known in finding solution to
such problem.
SVM is the binary classifier. As shown in Fig 5 it
classifies detected objects into two classes i.e. human and nonhuman. Hyperplane is constructed to differentiate between two
classes.
Fig 5 SVM Classifier
The line differentiates green and blue region is
hyperplane. The hyperplane is equidistant from two closest
2172
IJRITCC | April 2015, Available @ http://www.ijritcc.org
_______________________________________________________________________________________
International Journal on Recent and Innovation Trends in Computing and Communication
Volume: 3 Issue: 4
ISSN: 2321-8169
2171 – 2173
_______________________________________________________________________________________________
points from both the classes. Blue portion consist of nonhuman objects and green portion is for humans.
VII.
EXPERIMENTS RESULTS
We have tested the proposed human detection
technique on a number of videos. These videos are of various
real time scenarios.
Single human videos: Only single person appears. Videos have
different backgrounds.
Multiple human videos without occlusion: These videos have
multiple humans but they move in such a way that occlusion
doesn't occur. The study shown the human detection result
with video consisting of human and non-human objects at a
resolution of 720*1080, a frame rate of 22fps and for effective
calculation purpose we skipped 10 frames per second. The
video consisted of around 300-400 frames. One can observe
from Fig 5 that the proposed method classifies objects into two
classes from one considered frame accurately.
VIII.
SHIMOSAKODA, and Satoshi NAKAGAWA “Automated
Detection of Human for Visual Surveillance System” 1996 IEEE
Proceedings of ICPR '96
[9] Josef Grahn, “Using SVM for Efficient Detection of Human
Motion” Proceedings 2nd Joint IEEE International Workshop on
VS-PETS, Beijing, October 15-16, 2005
[10] Soumya Varma, Sreeraj M “Object Detection and Classification
in Surveillance
System” 2013 IEEE Recent Advances in
Intelligent Computational Systems (RAICS)
CONCLUSION
Nowadays various sectors are facing the security
problem hence we introduced this system so that human can
be detected in video surveillance system.
Input video is preprocessed i.e. converted into frames and
background subtraction is performed. Foreground objects are
obtained from background subtraction. Features are extracted
from these objects. These features are given as input to the
SVM classifier which classifies the object into Human or nonhuman objects
REFERENCES
[1] Manish Khare, Rajneesh Kumar Srivastava and Ashish Khare,
“Automatic
Multiple Human Detection and Tracking for
Visual Surveillance System”, IEEE/OSAIIAPR International
Conference on Informatics, Electronics & Vision
[2] Darin Brezeale and Diane J. Cook, Senior Member, IEEE
“Automatic Video Classification: A Survey of the Literature”,
IEEE TRANSACTIONS ON SYSTEMS, MAN, AND
CYBERNETICS, VOL. UNKNOWN, NO. UNKNOWN,
UNKNOWN 2007
[3] Ruofeng Tong, Di Xie, and Min Tang “Upper Body Human
Detection and Segmentation in Low Contrast Video” IEEE
TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR
VIDEO TECHNOLOGY, VOL. 23, NO. 9, SEPTEMBER 2013
[4] Hao Su1, Min Sun, Li Fei-Fei, Silvio Savarese “Learning a dense
multi-view representation for detection, viewpoint classification
and synthesis of object categories”
2009 IEEE 12th
International Conference on Computer Vision (ICCV) 978-14244-4419-9/09/$25.00 ©2009 IEEE
[5] Thomas MORANDUZZO and Farid MELGANI “A SIFT-SVM
METHOD FOR DETECTING CARS IN UAV IMAGES” 9781-4673-1159-5/12/$31.00 ©2012 IEEE
[6] Qiang Chen1, Zheng Song, Zhongyang Huang, Yang Hua and
Shuicheng Yan “Contextualizing Object Detection and
Classification” IEEE TRANSACTIONS ON PATTERN
ANALYSIS AND MACHINE INTELLIGENCE, VOL. XX,
NO. XX, 2013
[7] A. Ess1, B. Leibe, K. Schindler, L. van Gool “Moving Obstacle
Detection in Highly Dynamic Scenes” 2009 IEEE International
Conference on Robotics and Automation Kobe International
Conference Center Kobe, Japan, May 12-17, 2009
[8]
Yuji
KUNO,
Takahiro
WATANABE,
Yoshinori
2173
IJRITCC | April 2015, Available @ http://www.ijritcc.org
_______________________________________________________________________________________
