Inspection of Suspicious Human Activity in the Crowd Sourced Areas Captured in Surveillance Cameras

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 04 Issue: 12 | Dec-2017 www.irjet.net p-ISSN: 2395-0072
© 2017, IRJET | Impact Factor value: 6.171 | ISO 9001:2008 Certified Journal | Page 802
INSPECTION OF SUSPICIOUS HUMAN ACTIVITY IN THE
CROWDSOURCED AREAS CAPTURED IN SURVEILLANCE CAMERAS
P.Bhagya Divya1, S.Shalini2, R.Deepa3 , Baddeli Sravya Reddy4
1&4PG scholar, Department of Computer Science and Engineering,
2&3Assistant Professor, Department of Computer Science and Engineering,
1, 2,3&4 Prince Dr.K.Vasudevan College of Engineering and Technology, Chennai, Tamil Nadu, India.
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - The ultimate aim is to provide the indoor security
using the CCTV camera. The CCTV Camera is a video camera
that feeds or streams its image in real time; Webcams are
known for their low manufacturing cost and their high
flexibility, making them the lowest-cost form of video
conversations and inefficient security issues. The system will
detect suspicious person i.e. unauthorized entry in a restricted
place in a video by using AMD algorithm and will start
tracking once the user has specified a suspicious person by
his/her on the display. The main purpose of background
subtraction is to generate a reliable background model and
thus significantly improve the detection of moving objects.
Advanced Motion Detection (AMD) achieves complete
detection of moving objects. A camera is beenconnectedinside
the monitoring room which produces alert messages on the
account of any suspicious activity.
Keywords: CCTV camera, Advanced Motion Detection,
Background model, Suspicious activity, Webcams
1. INTRODUCTION
The region of the possibility and information
program focuses on research in various security techniques
to address problems in repeated aim detection applications.
The goal of computerized surveillance system is to support
the human operator in prospect investigation and event
categorization by without human intervention detectingthe
objects and analyzing their actions using computer vision,
techniques of pattern recognition and signalprocessing.This
review addresses more than a few advancements made in
these fields' while bringing out the detail that realizing a
practical end to end surveillance system still remnants a
hard job due to more than a few challenges faced in a real
world situation. With the improvement in computing
technology and now in-expensively and technically possible
to adopt multi camera and multi-modal structure to gather
the requirement of well-organized surveillance system in
broad range of security applications like security guard for
important buildings and surveillance in cities.
Visual surveillance has been an energetic study area in
pattern analysis and machine intelligence, due to its vital
position in helping surveillance intelligence and law
enforcement agencies to battle alongside offense and crime
actions. The objective of a visual surveillance system is to
identify irregular object behaviors and to lift alarms when
such behaviors are detected using the Advanced Motion
Detection (AMD) algorithm.
(x,y)=
After moving objects are detected, it is necessary to
categorize them into predefined categories, so that their
movements behaviors can be suitably interpret in the
background of their identitiesand theirconnectionswiththe
surface. Therefore, object categorization is a very important
part in a complete visual surveillance system.
1.1 Related work
The SNV technology presents a novelframeworkfor
recognize human activities from video sequences captured
by depth cameras. They extend the surface normal to
polynomial by assembling local neighbouringexcitedsurface
normal’s from a depth sequence to jointly characterize local
motion and shape information. Then they propose a general
scheme of Super Normal Vector (SNV) to cumulate the low-
level polynomials into a discriminative representation [1].
Binary range-sample feature in depth is implemented. The
goal is to engender front, activity, and back layers. Seeds for
generating the two bounding planes to separate them are
required. Joint points with depth less than zero can be
naturally regarded as the front seed points. This is a very
coarse operation, but is already sufficient in our feature
construction. It hasdenoted them respectively asCfrontand
Cback.[2]
A multi-part bag-of-poses approach is then defined, which
permits the separate alignment of body parts through a
nearest-neighbor classification. Experiments conducted on
the Florence 3D Action dataset and the MSR Daily Activities
dataset show promising results. This method has been
evaluated on two samples: the Florence 3D Action Dataset
and the Microsoft (MSR) Daily Activity 3D data set [4]. The
curiosity in the capturing of human actions is motivated by
the promise of many application, both offline and online.For
gradually more large and complex datasets, manual labeling
will become prohibitive. Automatic labeling using video
subtitles and movie scripts is possible in some domains, but
still requires manual verification. They discussed vision-
based human action recognition in this survey but a multi-
modal approach could improve detection in some domains

[5]. The temporal evolution of a modality appears to be well
approximated by a sequence of temporal segments called
onset, apex, and offset. The experimental results obtained
show the following: 1) affective face and body displays are
simultaneous but not strictly synchronous; 2) explicit
detection of the temporal phases can improve the accuracy
of affect recognition; 3) recognitionfromfusedfaceandbody
modalities performs better than that from the face or the
body modality alone; and 4) synchronized feature-level
fusion achieves better performance than decision-level
fusion.[6].
Trajectory captures the local motion information of the
video. A dense representation guarantees a goodcoverageof
foreground motion as well as of the surrounding context.
Additionally, they present a descriptor based on Motion
Boundary Histograms (MBH) which rely on differential
optical flow. The MBH descriptor shows to consistently do
better than other state-of-art descriptors, in particular on
real-world videos that contain a significant amount of
camera motion [9]. A filtering method is implemented to
extract STIPs from depth videos (called DSTIP) that
effectively suppress the noisy measurements. Further, they
have built a novel Depth Cuboids Similarity Feature (DCSF)
to describe the local 3D depth cuboids around the DSTIPs
with an adaptable supporting size. Experimental evaluation
shows that the proposed approach outperformsstate of-art
activity recognition algorithms on depth videos, and the
framework is more widely applicable than existing
approaches. They also givedetailed comparisons with other
features and analysis of choice of parameters as a guidance
for applications[10].
2. EXISTING SYSTEM
Existing approaches requires the user to record a
video at the faces and then process it to recognize them,
although the picture taken by user may not be able to
capture the image using the Depth cameras.
Fig 1: Detection of human actions by capturing the
images using the Depth cameras [Ref 1]
The depth images are in the appearance of the shaded
human figure as shown in the figure 1. They mainly
concentrate on the study or action recognitionofthehumans
which can be used for further examination on human
civilization. The use the technique of Super Normal Vectors
(SNV) and uses the implementation of polynormals.
U= → Equation of polynormal from
[1]
The images retrieved from the depth camerasimagescannot
be used for the identification of the human faces or some
other unique identification. Existing researches has a major
drawback of inefficiency in the case of online processing of
videos for crime reduction.
3. PROPOSED SCHEME
The CCTV Camera is a video camera that feeds or
streams its image in real time .The system will detect
suspicious person i.e. unauthorized entry in a restricted
place in a video by using AMD algorithm and will start
tracking once the user has specified a suspicious person by
his/her on the display. The main purpose of efficient
background subtraction method is to generate a reliable
background model and thus significantly improve the
detection of moving objects.
Advanced Motion Detection (AMD) achieves complete
detection of moving objects. A camera is been connected
inside the monitoring room which produces alert messages
on the account of any suspicious activity.
A. Background Modeling (BM)
Background subtraction, also known as foreground
detection, is a technique in the field so image processingand
computer vision where in an image's foregroundisextracted
for further processing (object acknowledgment, etc .).
(x,y)=
Fig 2: Advanced Background subtraction model

Generally an image's regions of interest are objects in its
forefront. The sample image that is included in the video for
detecting the moving objects cam be given as andthe
image taken for motion detection can be given as
(x,y).Then I(x,y) is the input video frames.
The architecture of the suspicious human activity detection
system is provided with seven modulesthat is includedwith
an alarm trigger system.
Fig 3: System Architecture for detecting the suspicious
human activities
B. Frame Sequence
The video to frame detection can be done by using so many
software’s which are available in the market today however
when we are using this software to get the frames from the
video software will decide in the begin itself how many
frames we need per second so which indicatesthattherewill
be a chances of missing the frames on which we are intent
more, normally the number frames per second will be
different for the different cameras.
Fig 4: Conversion of raw videos into frames
C. Object Extraction
A new method of video objectextractionisproposed
to accurately obtain the object of interest from actively
acquired videos.
Fig 5: Extraction of images of the humans from the
converted frame sequences
Traditional videoobject extraction techniques oftenoperate
under the assumption of standardized object motion and
extract various parts of the video that are motion consistent
as objects. In contrast, the proposed active video object
extraction (AVOE) paradigm assumes that the object of
interest is being actively tracked by a non-calibratedcamera
under general motion and classifies the possible actions of
the camera that result in the 2D motion pattern asrecovered
from the image succession.
The result that is obtained as a output of the Phase I work is
the identification of the human faces.It ca be given as the
result:
Fig 6: Phase I output screenshot

D. Detection of Suspicious activity
Detection of suspicious activity by video
surveillances is highly effective. In previous decade
monitoring of video by humans those are sited in front of
screen of videos captured by either CCTV or any other
cameras. Now we are going to automate this type of
monitoring the best techniques which is used by most cases
is image processing. Pattern analysis is a method of
surveillance specifically used for documenting or
understanding a subject's (or many subjects') behavior.
The system follows 3 main constraints such as , Height,Time
and body movement. When the constrains are satisfied for
the activities of a particular person, he will be considered as
a doubtful person to be reported.
E. Advanced motion detection (AMD) algorithm
Algorithm 1: Computation of human action
Input: a frame images I
a coding operator C
a activity D = (dk)K
k=1
a set of space-time cells I = fs
Output: Action Detection
1 computes polynormals fpig from I
2 compute coefficients f_ig of fpig by R
3 for space-time cell i = 1 to jI j do
4 for visual word k = 1 to K do
5 uki
: = spatial average pooling and temporal
Max pooling of _i;k (pi � dk), where pi 2 vi
6 end
7 Ui :=
u1i
; : : : ;uKi
_
8 end
4. CONCLUSIONS
The system has presented a novel module that generated an
accurate background with production of neither inefficient
pixels nor artificial “ghost” trails. After a high quality
background model wasproduced, the AT moduleeliminated
the unnecessary examination of the entire background
region and reduced the computational complexity for the
consequent motion detection phase. The proposed object
extraction module detected the pixel of moving objects
within the triggered alert region to from the moving object
mask. It also initiates the development of a system for
suspicious human monitoring and study of their behaviors.
Finally this algorithm works for Online (Real-time) video
processing and its computational involvedness is low.
In future, the system can be used with the highly accessible
storage service and it can also be implemented with hi-tech
mode of capturing of videos in the surveillance areas.
REFERENCES
[1] “Super Normal Vector for Human Activity
Recognition with Depth Cameras”, Xiaodong Yang,
Member, IEEE, and YingLi Tian, Senior Member,
IEEE
[2] C. Lu, J. Jia, and C. Tang, “Range-Sample Depth
Feature for Action Recognition”, CVPR, 2014.
[3] J. Shotton, A. Fitzgibbon, M. Cook, T. Sharp, M.
Finocchio, R.Moore, A. Kipman, and A. Blake, “Real-
Time Pose Recognition in Parts from Single Depth
Images”, CVPR, 2011.
[4] L. Seidenari, V. Varano, S. Berretti, A. Bimbo, and P.
Pala, “Recognizing Actions from Depth Cameras as
Weakly Aligned Multi-Part Bag-of-Poses”, CVPR
Workshop on Human Activity Understanding from
3D Data, 2013.
[5] R. Poppe, “A Survey on Vision based Human Action
Recognition”, Image and Vision Computing, 2010.
[6] H. Gunes and M. Piccardi, ”Automatic Temporal
Segment Detection and Affect Recognition from
Face and Body Display”, IEEETrans. Systems, Man,
and Cybernetics - Part B: Cybernetics, 2009.
[7] O. Oreifej and Z. Liu, ”HON4D: Histogram of
Oriented 4D Normals for Activity Recognition from
Depth Sequences”, CVPR, 2013.
[8] J. Luo, W. Wang, and H. Qi, ”Group Sparsity and
Geometry Constrained Dictionary Learning for
Action Recognition from Depth Maps”, ICCV, 2013.
[9] H.Wang, A. Klaser, C. Schmid, and C. Liu, ”Dense
Trajectories and Motion Boundary escriptors for
Action Recognition”, International Journal on
Computer Vision, 2013.
[10] L. Xia and J. Aggarwal,”Spatio-Temporal Depth
Cuboid Similarity Feature for Activity Recognition
Using Depth Camera”, CVPR, 2013.
[11] Advanced Motion Detectio(AMD) technique:
http://ieeexplore.ieee.org/abstract/document/560
5242/?reload=true

[12] video surveillance performance :
https://www.videosurveillance.com/cctv-
technology/cctv-video-management.asp
[13] Backgroung modelling strategy:
https://docs.opencv.org/3.2.0/d1/dc5/tutorial_bac
kground_subtraction.html
[14] Human Face identification and recognition:
https://facedetection.com/algorithms/
[15] Online video processing strategy:
https://online.duke.edu/course/image-video-
processing/

Inspection of Suspicious Human Activity in the Crowd Sourced Areas Captured in Surveillance Cameras

More Related Content

What's hot (15)

Similar to Inspection of Suspicious Human Activity in the Crowd Sourced Areas Captured in Surveillance Cameras (20)

More from IRJET Journal (20)

Recently uploaded (20)

Inspection of Suspicious Human Activity in the Crowd Sourced Areas Captured in Surveillance Cameras