Smart Security Surveillance

#1 #2 #3 #4
Khushi Udasi , Amarnath Chavan , Kaustubh Kashale , Omkar Gattawar , Shailesh Bendale *5
#
Dept. of Computer Science, NBN Sinhgad School Of Engineering Pune, Maharashtra

Abstract — The COVID-19 pandemic epidemic has A programming function library called OpenCV is primarily
paralysed the entire international health system. To stop the focused on real-time computer vision, machine learning, and
infection from spreading, it is now essential. Around the image processing. The primary objectives of computer vision
world, putting on a mask, washing our hands often, and are to manipulate and extract data from a real-time source. It is
avoiding social contact have all become priorities. The applicable to automobiles with autonomous driving. The
World Health Organization (WHO) strongly advises detection of faces, objects, and handwritings is done using
against spreading the new coronavirus by using a mask that OpenCV. It is crucial for both mask- and face-mask detection.
covers the mouth and nose. The primary objective of a
computer vision system is to find moving objects. The Many different moving object identification methods have
performance of these systems is insufficient for many been described in recent years. In contrast to human vision, a
applications. One of the main causes is that moving object machine vision system's artificial method of detecting motion
detection is more difficult when dealing with various was based on statistical backdrop modelling. Our strategy was
constraints, including environmental changes. After object to periodically take a photo using the camera (make it the
recognition, object counting is a process that is improved current picture), compare it to a prior picture, and keep both if
upon and made more reliable with the aid of OpenCV. there was a noticeable difference; otherwise, we would release
OpenCV comes with a variety of helpful methods for item memory from the old picture and make the new picture the
recognition and counting. Numerous industries, including current picture.
transportation, health care, and environmental science, use Motion detection encompasses the recognition and
object counting. documentation of the complete movement procedure.
Automatically detecting movement in a certain area is the aim
Keywords: Face Mask Detection, Motion Detection, Object of motion detection [3]. Finding movement in a certain region
Recognition, OpenCV, Image Processing. is the aim of motion detection. In camera geometry, this area is
always represented by a zone of awareness known as the field
I. INTRODUCTION of vision. It is also known as an environment that is monitored.
With the growth of technology, there is a growing worry for In this instance, the environment with its moving objects and
safety and security everywhere nowadays. The number of activity is the focus. A zone of interest, which is referred to as
security cameras has lately grown in order to address this a moving object, might include a person, an animal, or an
problem. Despite this, it is challenging to manually store and object.
monitor data [1] continuously. Unexpected effects of the
coronavirus vary among people. Mild to severe sickness will be
experienced by the majority of infected persons. and recuperate II. RELATED REVIEW ON LITERATURE
outside of the hospital. Fever, a dry cough, and fatigue top the Many different moving object identification methods have
list of usual symptoms[2]. Therefore, using masks and hand been described in recent years. A strategy based on statistical
sanitizers has proven to be successful in preventing the background modelling was recommended by Mahbub et al. [4].
transmission of the virus. As a result, a face mask detection To find moving objects, this technique compares each edge
system is needed, which will inform the public and aid in the segment of the current frame with each edge segment of the
fight against the epidemic. In this research, we investigate real- backdrop. However, this method is unable to identify a moving
time face mask detection with deep learning and OpenCV. edge segment that crosses a background edge segment. The
Deep learning is a branch of machine learning that uses goal of Geetha Priya S et al[5] .'s You Only Look Once (YOLO)
algorithms with human brain-inspired design. In many approach for object detection is to find things. This
domains, deep learning aids in the breakdown of issues. methodology offers a lot of benefits over other object detecting
Convolutional neural networks, image detection, and picture strategies. While other algorithms, like Convolutional Neural
classification are all provided by deep learning (CNN). Network and Fast Convolutional Neural Network, only
Computer detection and classification tasks primarily utilise partially examine the image, YOLO does so by using
convolutional neural networks (CNN). In this study, deep convolutional networks to predict the bounding boxes and the
learning algorithms are utilised to distinguish between faces class probabilities for these boxes. As a result, YOLO detects
wearing masks and those who are not. To incorporate the the image more quickly than other algorithms.
optimal amount of Convolutional Neural Layers for precise The authors demonstrated the results of adding several filters to
detection, Convolutional Neural Networks (CNN) are utilised. a picture, such as colour conversion, grey scale, and dilation.
To gauge how long the algorithms required to finish, the
authors tested the features on three distinct smartphones. An
instructional manual on how to use OpenCV for image
recognition was written by the authors in [6].

For crucial industries like banks and enterprises, the authors

want to create a good motion detection system. In order to
determine the difference between the sampled photos, they start
by extracting samples from a webcam that is capturing images.
A counter is increased when they detect movement, and when
it reaches a certain level, they send a message to a mobile phone
informing it that movement has taken place and sound a buzzer.

The Gaussian Mixture Model, a probability density function,

served as the foundation for the authors' strategy to achieve this
(GMM). The scientists changed the Gaussian parameters and
component count of the traditional GMM method to shorten
processing time. Each pixel in each frame was subjected to the Figure 1 The Relationship between Computer Vision &
authors' GMM method to create foreground-background AI
segmentation. Once the foreground and background have been
defined, the authors convert the frame to binary, with the 3. 2 Image processing
foreground turning white (moving things) and the background
turning black (none moving objects). The binary pictures are Computer vision is not the same as image processing. Image
then subjected to filters by the authors to remove noisy processing is the act of transforming an existing image into a
elements as wind, light, and shadows. new one , typically by streamlining or improving the
information. It is a type of digital signal processing that is not
focused on deciphering the content of the images. For some
computer vision systems, image processing, such as picture
III. Computer Vision pre-processing, may be necessary.
Motion detection can be used to automatically carry out a job
when motion is detected. Motion detection, for instance, can be
Computer vision (CV) is a subfield of computer science that used in security systems to identify illegal activity or to
focuses on giving computers the ability to understand visual illuminate a room when people enter it.
data. Gerald Jay Sussman was given the task by Martin Minsky
in the early 1970s or late 1960s to connect a computer to a Human bodies emit infrared radiation when they produce heat,
camera and have the device report what it saw. which an infrared sensor can detect. Image processing can
identify motion by comparing two pictures [6] to one another.
The field of research known as "computer vision" (CV) strives By comparing pixels in the same positions in the two
to develop methods that will enable computers to "see" and photographs, this is done. There is no difference between two
decipher the content of digital images such as photos and pixels if the photographs are identical. The pixel difference
videos. It appears to be simple since everyone, even very young between certain pixels will be more than zero if the photographs
toddlers, can figure out the computer vision issue. Despite this, are not identical. Differences in the pixels where the people are
the issue is still largely unresolved due to both a lack of in the image will be visible when comparing an image of a room
knowledge on biological vision and the intricacy of visual that is empty to an image of the same room with people inside.
perception in a dynamic physical universe that is practically
infinitely changing.

3.1 Relationship between computer vision and AI IV. OpenCV

As seen in figure 1, computer vision is a field of study that The infrared radiation that human bodies release is detected
focuses on helping computers see. It is a multidisciplinary using an infrared sensor. A free software library for computer
discipline that falls under the umbrella of machine learning and vision and machine learning is called OpenCV (Open Source
artificial intelligence, and it may make use of both specialised Computer Vision Library). Computer vision is essential to
methods and generic learning algorithms. artificial intelligence or serves as its main foundation.
Computer vision is vital to robots, self-driving automobiles,
and image editing software. For taking, processing, and
analysing photos, an Android application subsystem called
OpenCV was used.
Additionally, these algorithms have MATLAB, Java, and
Python bindings. It primarily concentrates on face
VI. Existing Problem
identification, image processing, and video capture. The
hardware acceleration of the underlying heterogeneous This paper's major objective is to enable a basic camera
computing architecture benefits Open CV. Nearly 2,500 attached to a general-purpose computer to identify objects
algorithms for various computer vision methods are available moving through it and to determine how long the object
through Open CV. Because Open CV is a cross-platform remained in the region where the camera is placed.
library, it can run on any OS. Consequently, this programme may be used for monitoring.
Figure 2 illustrates how to take the first frame using a
The library, which comprises a thorough blend of both camera. This frame will serve as the initial reference. To
traditional and cutting-edge computer vision and machine detect motion, the phase difference between the new frame
learning approaches, has more than 2500 optimised algorithms. with the item and the baseline frame will be computed. The
For the purpose of processing images, OpenCV was developed. brand-new frames will be called Delta frames. Your delta
Each function and data structure was created with the image frame will then be adjusted using pixel intensity. The
processing programmer in mind. Contrarily, Matlab is a fairly improved frame will be called Threshold. Then, to capture
open-ended programming language. We can get just about significant elements on the Threshold frame, you'll employ
anything in the world in the form of toolboxes. To do video complex image processing techniques like Shadow
tracking, an algorithm analyses successive video frames and Removal, Dilation, Contouring, and others. Here is an
outputs the movement of targets between the frames. illustration of what you will achieve. The timestamps of
objects entering and leaving the frame can be recorded.
V. OpenCV You'll be able to calculate the duration of screen time as a
consequence.
The detection of a face mask may be broken down into the
following two steps: Face recognition and mask detection are
the first two. Viola-Jones method, a machine learning object
identification approach, and Haar feature-based cascade
classifiers are used with OpenCV to accomplish face mask
recognition. Rapid Object Detection Using a Boosted Cascade
of Simple Features, a 2001 study by Paul Viola and Michael
Jones, introduced the face detection technique. We must carry
out face detection for each frame of the video in order to
achieve real-time face identification in a video. In figure2,
face detection is displayed.

The image is now preprocessed and reduced in size to

224x224 pixels when the face is detected in it. Additionally,
OpenCV spretrainedHaar course classifiers are used to
identify the mouth and nose, which determines if a person is
wearing a cover. If the person's lips and nose can be clearly
seen, they are not wearing a veil, hence it is not certain Figure 3. The Detected Object
whether they are covered.
VII. Proposed Methodology

In our project, we aimed to develop a surveillance system

that, in addition to motion detection, could alert the user of an
incursion, record video from the instant the motion was
detected, and send SMS messages to the user's mobile device.

7.1 Comparing phase

We compare the webcam's live photos to one another in order

to detect changes in these frames and predict the presence of
motion in order to determine whether any motion is present in
the live Images.

Figure 2: Face detection.

 at the same time, an SMS is delivered to the user's mobile
device.

VIII. Object Detection

Every part of an object throws a shadow on the surrounding

area or on other parts of the object. This can be confusing. For
instance, the nose throws a shadow on your lips, any bigger
stationary item casts a shadow on smaller surrounding objects,
various light sources with variable luminous intensities, your
room's curtains, the direction and viewing angle of the light
Figure 4. Flow Diagram
source, etc. Some of the irregularities that have been identified
in real-time collected frames are as follows.
7.2 Image segmentation We must filter the image as a consequence to lessen these types
of disruptions. By altering the number of iterations, we may
The process of splitting a picture into parts, each with their influence how smooth the Dilate function is.
unique set of attributes, in order to extract the target of interest Using the imshow function, which lets you display each frame
is referred to as "image segmentation." The image segmentation in an own window, you may compare frames.
technique used in this work is threshold segmentation. To put it
another way, the foreground and background of an image are
divided into two groups depending on the results of comparing
a range in the picture to the threshold. The threshold
segmentation procedure consists of two basic.
If we can locate a suitable threshold, we can modify the image
for segmentation. An intelligent video surveillance system may
be used to identify moving targets in a variety of different ways,
such as frame difference and background removal techniques.
Threshold segmentation is composed of two basic procedures.
First, establish the threshold. The second step, which is the most
crucial one in partition, compares the pixel value to the
threshold value T to determine the threshold value.
Using the difference between the current picture and the
background image, the background subtraction method is a
method for identifying moving objects.
After subtracting the current image fk from the previously
saved background image B, if the pixel difference above the
bound threshold, the background pixel or the pixel to pixel on
the moving target is identified. The background subtraction
threshold must be properly selected in order to successfully Figure 5. The Frame with Detected Object
identify motion. Motion detection's precision is crucial.
We use the waitKey function to postpone the operation till a
7.3 System indicating phase key is pushed. In this instance, waitKey is used to obtain an
ongoing live feed from the camera (1). To prevent the video
from being recorded, just press the "Q" key on your keyboard.
The program contains an alarm mechanism because when
motion is detected, the user may wish to get an SMS alert
immediately away informing them that the software has IX. Results and Discussion
detected an intrusion. This alarm system quickly creates an
9.1 Motion detection
auditory alert signal in wav file format whenever any type of
motion is detected. This helps at the moment to avoid any A home or business security system that may be very helpful
in circumstances when security is a worry is the "Smart
kind of security compromise. When motion is detected, the webcam motion detection surveillance system." The
user is provided a location with photos of the trespasser, and development of technology in the modern world has
significantly increased the techniques employed by thieves and
criminals to steal. As a result, surveillance systems need to
develop in order to keep up with the world's changing needs.
The most current tools used in the war against theft and
devastation are video surveillance and monitoring.
The most important feature of digital video surveillance
systems is motion detection. It enables the camera to just
capture when necessary rather than constantly, which
significantly reduces storage space. An alert may sound when
unexpected motion is discovered. As a result, personnel are no
longer necessary to constantly monitor. The motion detector
satisfies the need for an inexpensive, minimal security system
in daily life. The future holds great potential for computerised
home security. Future is bright and made simpler with new
technology.

Figure 6. The Output Format

9.2 Mask detection

Using the Python computer language, real-time face mask

identification with 98% validation accuracy was implemented. Figure 8: Testing results of mask detection.
After a number of testing with a batch size of 32 and 20 repeats
of the epoch, this is the rate that stands out the most. The results X. Conclusion And Future Work
show how well the trained model performs in terms of accuracy
and loss. The most modern technologies used in the struggle against
theft and destruction are video monitoring and surveillance.
With the help of technology, it is possible to see and record
every square inch and passing second of the area of interest.
The most important feature of digital video surveillance
systems is motion detection. It enables the camera to just
capture when necessary rather than always, which significantly
reduces storage space. An alert may sound when unexpected
motion is discovered. This frees up staff from ongoing
supervision. The motion detector satisfies the need for an
affordable, basic security system in daily life. The future of
computerized home security is quite promising. With new
technology, the future appears promising and simpler.
This application may be used for surveillance to secure any
Figure 7: Training loss and accuracy during model location or significant item from illegal access and to track how
training. long an unauthorized person or object tries to breach the
perimeter or enter the protected region. Additionally, more
beneficial techniques and algorithms might be used to enhance
the processes, such as those that track the object that generates
the motion events or those that employ optical flows, also
known as image flows. There are indications of human
movements here. However, in the future, it could be desirable
to recognize the human body posture or gesture that the
prototype system's algorithm captured.
 Based on the findings of the study presented in this article, XII. REFERENCES
real-time face mask identification using OpenCV and deep
learning algorithms is a remarkable solution for straightforward
[1] Real-Time Face Mask Detection using OpenCV and
facemask recognition. There are, however, a relatively small DeepLearning, Department of ECE, KoneruLakshmaiah
number of restrictions that may be readily solved in future Education Foundation, Andhra Pradesh, India.
work. The suggested approach is especially helpful for real-
time applications that need face mask recognition, particularly [2 ] An Intelligent Motion Detection Using OpenCV.
in heavily crowded areas like schools, airports, and public International Journal of Scientific Research in Science,
areas. utilising a variety of IOT applications and deep learning Engineering and Technology Print ISSN: 2395-1990
approaches.
We can also use contactless temperature monitoring and
mask detection, which can determine whether the mask is being [3] Y. You, S. Gong, C. Liu, "Adaptive moving object
worn properly and grant access only if both factors are met. If detection algorithm based on back ground subtraction
neither of these conditions is met, or if the body temperature is and motion estimation", Int. J. Advancements in
above normal, they can alert us, allowing us to take the Computing Technology, vol. 5, no. 6, pp. 357-363,
necessary precautions to stop the spread of the Covid-19 virus
indoors. 2013 Conference,WWW2019,2019.

[4] 9M. Murshed, A. Ramirez, O. Chae, “Statistical

XI. ACKNOWLEDGMENTS
Background Modeling: An Edge Segment Based
Moving Object Detection Approach”, Proc. of IEEE
Our first and foremost acknowledgment is to our supervisor
and guide Prof. Shailesh P. Bendale. During the long journey International Conf. on Advanced Video and Signal
of this study, he supported me in every aspect. He was the one Based Surveillance, pp. 300-305, 2010
who helped and motivated us to propose re-search in this field
and inspired me with his/her enthusiasm on research, his
experience, and his lively character.
[5] Geethapriya. S, N. Duraimurugan, S.P.
we express true sense of gratitude to our guide Prof. Shailesh
P. Bendale for his perfect valuable guidance, all the time Chokkalingam, “Real-Time Object Detection with
support and encouragement that he gave us. Yolo”, International Journal of Engineering and
We would also like to thanks our head of department Prof. Advanced Technology (IJEAT) ISSN: 2249 – 8958,
Shailesh P. Bendale, principal and management inspiring us
and providing all lab and other facilities, which made this
Volume-8, Issue-3S, February 2019.
seminar presentation very convenient.
We are really thankful to all those who rendered their valuable [6 ] Marengoni and Stringhini. High Level Computer
help for the successful completion on the seminar presentation. Vision using OpenCV. 2011. Universidade
Presbiteriana Mackenzie.