Detecting people not following covid norms and masks over face

A
Project Report
Submitted for the partial
fulfillment
of B.Tech. Degree
in
COMPUTER SCIENCE & ENGINEERING
by
Navin Chandra (1805210027)
Arnab Debnath (1805210011)
Tushar Giri (1805210059)

Under the supervision of

Mr. Abhishek Singh
Dr. Pawan Kumar Tiwari

Department of Computer Science and Engineering

Institute of Engineering and Technology

Dr. A.P.J. Abdul Kalam Technical University, Lucknow, Uttar Pradesh.
July 2022
 Contents

DECLARATION.................................................................................................................i
CERTIFICATE...................................................................................................................ii
ACKNOWLEDGEMENT..................................................................................................iii
ABSTRACT.......................................................................................................................iv

LIST OF FIGURES............................................................................................................v

1. INTRODUCTION..........................................................................................................8

2. LITERATURE REVIEW.............................................................................................11
3. METHODOLOGY.......................................................................................................14
3.1 EXISTING SYSTEM...................................................................................14
3.2 PROPOSED SYSTEM.................................................................................14
3.3 HARDWARE REQUIREMENTS..........................................................................14
3.4 SOFTWARE REQUIREMENTS...........................................................................15
3.5 SYSTEM ARCHITECTURE.................................................................................15
3.6. MODULE DESCRIPTION.................................................................................... 15

3.6.1 IMAGE PREPROCESSING........................................................................................................16

3.6.2 DATA AUGMENTATION.........................................................................................................17

3.6.3 MODEL TRAINING....................................................................................................................18

3.6.4 TESTING THE MODEL.............................................................................................................18

3.6.5 IMAGE SEGMENTATION USING MASK CNN.....................................................................19

3.6.6.IMPLEMENTING THE MODEL IN OPENCV..........................................................................19

3.6.7 IMPLEMENTING FACE RECOGNITION................................................................................ 20

3.6.8 SEND SMS AND EMAIL TO A PERSON WHO IS NOT WEARING A MASK.....................22

4. EXPERIMENTAL RESULTS...................................................................................... 23
5. CONCLUSIONS...........................................................................................................27
REFERENCES
ANNEXURE

2
 List Of Figures

Fig 3.1:- Testing the Model Evaluation Metrics…............................................................19

Fig 3.2:- Flow chart for implementation of facial recognition...........................................22

Fig 3.3:- SMTP protocol.....................................................................................................23

Fig 4.1:- Testing the Model…............................................................................................24

Fig 4.2:- Detecting people wearing masks..........................................................................25

Fig 4.3:- Sending text messages to people not wearing masks.........................................26

Fig 4.4:- Email template sent to people….........................................................................27

3
 Declaration

We hereby declare that this submission is our own work and that, to the best of our belief and
knowledge, it contains no material previously published or written by another person or material
which to a substantial error has been accepted for the award of any degree or diploma of
university or other institutes of higher learning, except where the acknowledgment has been made
in the text. The project has not been submitted by us at any other institute for the requirement of
any other degree.

Submitted by: - Date: 26-05-2022

(1) Name: Navin Chandra

Roll No.: 1805210027
Branch: Computer Science and Engineering
Signature:
(2) Name: Arnab Debnath
Roll No.: 1805210011
Branch: Computer Science and Engineering
Signature:
(3) Name: Tushar Giri
Roll No.: 1805210059
Branch: Computer Science and Engineering
Signature:

4
 Certificate

This is to certify that the project report entitled “Recognizing people who aren’t wearing masks”
presented by Navin Chandra, Arnab Debnath, and Tushar Giri in the partial fulfillment for the
award of Bachelor of Technology in Computer Science and Engineering, is a record of work
carried out by them under my supervision and guidance at the Department of Computer Science
and Engineering at Institute of Engineering and Technology, Lucknow.

It is also certified that this project has not been submitted to any other institute for the award of
any other degrees to the best of my knowledge.

Mr. Abhishek Singh Dr. Pawan Kumar Tiwari

Department of Computer Science and Engineering
Institute of Engineering and Technology, Lucknow

5
 Acknowledgment

We would like to express our gratitude to Dr. Divakar Singh Yadav, Head, CSE Department, IET,
Lucknow, and our humble Supervisors Dr. Pawan Kumar Tiwari and Mr. Abhishek Singh for
their continuous support and suggestions so far in this project. They have been very constructive,
supportive, kind and made it easier to write up this thesis work. Their continuous support gave us
the motivation to work on this project and helped us to gain knowledge in various fields. I would
also like to thank them for suggesting changes. They allowed us to locate areas of improvement
that helped me to write this interim report.

Navin Chandra (1805210027):

Tushar Giri(18015210059):

Arnab Debnath(1805210011):

6
 Abstract

Our earth is going through a pandemic phase where people are getting affected by the COVID-19
virus and its different variants. This has resulted in a large number of casualties and security
concerns. People have been instructed to wear masks whenever they are in public places to
prevent the rapid spread of this virus. Therefore, we designed a model which will help us to make
people aware of the importance of wearing masks. The primary goal was to come up with an idea
that can motivate people to wear masks in public places by detecting the people who have not
worn masks. In this report, in order to address the issue of the masked face region, we propose a
model based on discarding the masked region and deep learning-based features. The first step is to
detect the people who are not wearing masks by discarding the masked faces. For this, we used
CNN. Then for recognizing the people who have not worn masks we used the facial recognition
algorithm which will help us find the details of those people. For awarding those people we will
be sending a precautionary message to wear masks in public places. All these can be observed
through live streaming from cameras at various public places like malls, markets, classrooms, etc.

7
 Chapter 1
Introduction

1.1 OVERVIEW

It has been almost 3 years that we are facing the wrath of COVID-19 all over the world affecting
the lives of many people. India is also not far away from this. India recorded one of the worst
effects of COVID leading to the economical crisis. Lots of people lost their lives. Lockdown was
imposed leading to strain on the world economy. After the two waves got over, relaxations were
given. WHO gave instructions on how to take care of this virus. They said that mask is a great
weapon to keep coronavirus away. Proper sanitization should be maintained whenever people go
out. People after getting relaxed in covid regulations started going out. Often we saw people not
maintaining social distancing when going out in public places. As we know how important it is to
maintain social distancing, sanitization, and wearing masks when we go out in public. After the
revocation of covid rules, we have seen people going out in crowded places without masks and
maintaining social distance. Surveys tell that almost 90% of people are aware whereas only 40
percent of people wear masks. From these surveys, we will say that people often neglect to wear
masks either because they are not comfortable or they don’t believe that there is such a thing
called covid. This kind of behavior by people leads to the quick spread of covid viruses rapidly
affecting the lives of many people. The WHO has thus said that only wearing masks and
maintaining social distancing is the key to keeping away the virus. The areas where people are
freely moving without masks are the key culprits. Also, the people living in rural areas where
there is no proper awareness, need to be aware of these facts and tips to keep away these viruses.
This uncertain outbreak of the COVID-19 virus has led us to the importance of wearing masks in
public places and keeping us safe from such viruses.

1.2 OBJECTIVE

The objective of this model is to design and develop a model that helps us to identify those people
who aren’t wearing masks either due to lack of awareness or negligence. It is created by the use of
MobileNet V2, Face Recognition Library, and Twilio.

8
1.3 PROBLEM STATEMENT

Nowadays, carrying masks is a critical want in public places, because, to a large extent, it will
become hard for protection officers to check each and every individual who is no longer wearing
a mask. Therefore, it is important to identify those people and make them aware of the importance
of wearing masks.

1.4 PROJECT DESCRIPTION

The MobileNet V2 algorithm will be used to create face masks in this project. We'll be training
our mannequin using images of people wearing masks and those who aren't wearing masks, which
will be sent as an entered dataset, and the segmented photo of the equation will be purchased as
an output. The most important phase in our model is pre-processing, which involves resizing the
image to a particular size and converting it to a NumPy array. One warm encoding is applied to
the pictures. Finally, the data are broken down into two categories: checking out and training. The
second phase is data enhancement, which involves flipping, zooming, and finally flipping the shot
horizontally. The third stage is training; the bottom model's region is loaded with imageNet
weights, and the closing layer of the pre-trained model is fine-tuned as a result. The usual pooling,
flattening, dense, activation function (relu and softmax), and the dropout rate are all entered and
processed inside the closing totally associated layer. The model summary is then obtained, and the
configuration is also recorded. The optimizer loss entropy and accuracy metric are specified in the
fourth phase, and the trained model is then assessed and stored. The schooling loss and training
accuracy data are plotted inside the fifth phase. Following that, a prediction on the teaching set is
completed. Finally, the model is put to the test. Image segmentation is the sixth phase, in which
the dataset's proximity is imported, ID mapping is complete in the region of the masks, and no
masks are issued. Following that, a TensorFlow session is created, and the Mask RCNN model is
loaded. After that, appropriate detection of Boxes, Classes, Scores, and Masks is completed. The
results of the detection are then shown after event segmentation. Finally, the mannequin is created
with the use of a webcam, which allows the video to be viewed through the body and scaled as
needed. The preprocessing feature is then used to determine the impacts of persons wearing
masks and not

9
wearing masks collectively, with a % accuracy. Then we identify those of us who aren't wearing
masks and match them with real individuals in our database to gain access to their contact
information. For this, we used a biometric identification library to match that person and collect
their information. Then, using an API request, we'll utilize Twilio to deliver messages and mail to
these men and women who are no longer wearing masks.

1.5 MOTIVATION

The world has been turned upside down over the past two years, owing to the very fatal virus Covid-
19. The spread of covid is still visible in most areas, with new varieties of the same virus-like
Omicron emerging. The mask is one item that must be worn frequently when going out these
days. However, many individuals are ignorant of genuineness and refuse/avoid putting on a mask.
This exercise aids in identifying those who do not appear to be wearing a mask. Masks have been
the most recent addition to a list of common items that includes wallets and purses.

10
 Chapter 2
Literature
Review

The Literature Survey is hired to deliver a quick evaluation and explanation of the reference
papers. The literature survey conveys the technical important points associated with the task in an
instead ideal and precise manner. Xinbei Jiang, Tianshan Gao, Zichen Zhu, and Yukang Zhao [1]
proposed a real-time masks detection tool with the usage of YOLOv3. This paper makes use of
the Masked Face Detection (PWMFD) dataset containing 9205 sports activities masks image
samples. Relationships among channels are executed via means of integrating Darknet53's eye
mechanism with the usage of SE blocks to permit the network to goal this feature. A Glou loss is
carried out to account for spatial variations among scoring and floor packing containers and to
enhance the robustness of bounding box regression. Excessive foreground/history kind imbalance
is dealt with as a lack of focus. The relaxation of the outcomes showed that SEYOLOv3 changed
into more healthy than YOLOv3 and different current PWMFD detectors. When evaluated with
YOLOv3, the proposed dummy confirmed an 8.6% extra mAP and detection rate. Samuel Ady
Sanjaya and Suryo Adi Rakhmawan [2] advanced mask detection with the usage of MobileNetV2.
Within the article, the laptop learns the MobilenetV2 set of rules for mask identification. The
steps concerned in the manikin introduction consist of facts gathering, pre-processing, facts
separation, version testing, and version application. The proposed dummy can offer an accuracy
of 96.85%. Sunil Singh, Umang Ahuja, Munish Kumar, Krishna Kumar, and Monika Sachdeva
[3] proposed a mask detection tool for the usage of YOLOv3 and quicker RCNN models. This
newspaper draws the eye to packaging packing containers whether or not or now no longer the
humans on the window in purple or green sun sunglasses are carrying masks, persevering with the
percentage of individuals who put on masks in regular life.
G. Jignesh Chowdary, Narinder Singh Puni, Sanjay Kumar Sonbhadra, and Sonali Agarwal [4]
developed a mask sensing device using InceptionV3 transfer learning. Inside the paper, a switch-
controlled dummy has been proposed to automate the approach of detecting a person not wearing
a sports mask. The dummy uses deep knowledge of the Inception V3 algorithm to observe the
mask. The Simulated Masked Face Dataset is used for training and testing. Due to constrained
availability, the picture graph growth approach is used for better training and version validation.
The dummy became 99.9% curate for the duration of schooling and 100% curate for the
duration of testing.
11
Shilpa Sethi, Mamta Kathuria, and Trilok Kaushik [5] carried out masks to hit upon using deep
mastering. To gain immoderate precision and brief inference time, the proposed approach makes
use of a single-degree detector and a -degree detector. ResNet50 and furthermore, sooner or later
in this article, the idea of a switcher mastering high-degree semantic statistics comes into play.
During mask detection, for you to improve localization performance, bounding area
transformation is hired. Three well-known baseline models viz. ResNet50, AlexNet, and
MobileNet are used for experimenting with the version. The proposed together with those models
can produce immoderate accuracy in a whole lot much less inference time. The proposed
technique 6 achieved an accuracy of 98.2% while carried out with ResNet50. in evaluation with
the presently published Retina facemask detector, the proposed model achieves 11.07% and
6.44% extra precision and bear in mind in mask detection. The proposed model is fantastically
prepared for video surveillance devices. Riya Chiragkumar Shah and Rutva Jignesh Shah [6]
proposed a tool for the Detection of Face masks with the usage of a Convolutional Neural
Network. The model proposed properly right here is supposed to be modeled using python
libraries, specifically TensorFlow, Keras, and OpenCV. The model used is the MobileNetV2 of
convolutional neural networks. Throughout this paper, a model is advanced using the above-stated
libraries. The model is tested for a variety of prerequisites with distinct hyperparameters. the
maximum critical dataset is fed into the version, and run by the academic program, which trains
the model at the given dataset. Then the detection utility is run, which activates the video stream,
and captures the frames continually from the video waft with an anchor area using an item
detection process. The output is then equipped MobileNetV2 layers the vicinity it's far labeled
into humans wearing a mask surrounded thru green containers and different people now not
carrying an area surrounded thru red boxes. Safa Teboulbi, Seifeddine Messaoud, Mohamed Ali
Hajjaji, and Abdellatif Mtibaa [7] advanced a tool in Real-Time Implementation of AI-Based
Face mask Detection and Social Distancing measuring gadget for COVID-19 Prevention. This
research paper makes a specialty of imposing a mask and Social Distancing Detection model as an
embedded vision system. The pre-educated models just like the MobileNet, ResNet Classifier, and
VGG are used. This paper includes fundamental blocks. The important block includes the 7
training and moreover the attempting out models, while the second block includes the whole
framework testing. This give-up end result detects people carrying a mask and now not wearing a
mask and guarantees social distancing. Xueping Su, Meng Gao, Jie Ren, Yunhong Li, Mian
Dong, and Xi Liu [8] carried out mask detection and class thru deep transfer

12
mastering. This paper describes an enterprise new set of rules for mask detection that integrates
transfer learning and Efficient-Yolov3, using EfficientNet because of the truth of the backbone
feature extraction network, and also you because of the truth of the loss function to restrict the
number of network parameters and beautify the accuracy of mask detection. This paper divides
the mask into instructions for licensed masks and unqualified masks create a mask class statistics
set and advise an enterprise new mask class set of rules then combines transfer reading and
MobileNet, improves the generalization of the version, and solves the rely on small information
size and clean overfitting. Mohamed Almghraby and Abdelrady Okasha Elnady [9] proposed a
tool for mask Detection in Real-Time using MobileNetv2. The created model for detecting face
masks at some stage in this paper uses deep mastering, TensorFlow, Keras, and OpenCV. The
MobilenetV2 set of rules is hired sooner or later in this paper to find out the face mask. this model
dedicates 80 percent of the education dataset to education and 20% to testing and splits the
training dataset into 80% education and 20% validation, primary to a final model with sixty 5
percent of the dataset for schooling, 15 percent for validation, and 20% for testing. Stochastic
Gradient Descent (SGD) is hired as an optimization approach with a getting-to-know rate of 0.001
and momentum of 0.85. 8 Chhaya Gupta and Nasib Singh Gill [10] proposed a tool of Corona
masks: A mask Detector for Real-Time Data. Convolutional Neural Network (CNN) set of rules is
hired sooner or later of this challenge to erect faces. During this paper, a dataset has been created
which includes 1238 pictures which might be divided into lessons “masks” and “no masks”. Live
streaming films may also moreover be used as input and extraordinary people wearing a mask and
now not wearing a mask may also be detected. The convolutional neural network is
knowledgeable about the dataset and it gives 95% of accuracy.

2.1 SUMMARY OF THE SURVEY

With Covid-19 cases on the rise all over the world, it's more important than ever to follow the
guidelines provided by the WHO and local health authorities. The above-mentioned document
was useful in visualizing and perceiving the overall project's flow, as well as defining its
architecture, input, and layout to be provided for each module, as well as identifying the expected
output. The above-mentioned articles are focused on recognizing persons who aren't wearing a
mask. In light of all of these articles, our project is also focused on using MobileNetV2 to detect
people who aren't wearing face masks.

13
 Chapter 3
Methodology

SYSTEM DESIGN
Software design should be a method for converting requirements into the appropriate
representation on a regular basis. This graphic will be used to construct the structure and targeted
plan diagrams, as well as the mission method.

3.1 EXISTING SYSTEM

Yolov3 is now being used to construct a system that is aware of face masks. They introduced a
spine community that will distribute more resources inside the current system, and they used Glou
and focus loss to speed up the coaching process and improve performance. It has an accuracy rate
of 86 percent. When using the appliance with one-of-a-kind algorithms, the accuracy will be
doubled.

3.2 PROPOSED SYSTEM

The mask detection in the proposed machine has been done using MobileNet V2. In comparison
to the previous system, MobileNet V2 will be able to detect face masks in a large group of men
and women and provide greater accuracy. As a result of the entry dataset and therefore the
segmented photo of the equation being received as output, provide an image of some persons
wearing masks and now without wearing masks. The mannequin is then performed employing a
camera, with the video being checked by way of body and scaled as needed. The preprocessing
feature is then used to encourage the impacts of those wearing masks and those who aren't.

3.3 HARDWARE REQUIREMENTS

The following hardware specifications are required to run this model:

● RAM – 8 GB

● Operating System(OS) – Windows 10

● Processor – Intel Core i5

14
 ● Processor speed – 3.60 GHz

3.4 SOFTWARE REQUIREMENTS

The Jupyter Notebook IDE and the Python programming language were used to create this
application.

● Programming Language – Python

● Python IDE – Sublime Text

● Deep Learning Framework - Tensorflow.

3.5 SYSTEM ARCHITECTURE

The structural plan denotes the many methods that will be employed in the job, while the exact
layout denotes the whole operation of each module that can be used for better perception and
execution of the project to achieve the desired results.

3.6 MODULE DESCRIPTION

The system can be divided into six modules.

Module1: Image Preprocessing

Module2: Data Augmentation

Module3: Model Training

Module4: Testing the Model

Module5: Image Segmentation using Mask CNN

Module6: Implementing the Model in Opencv.

15
Module7: Implementing Face Recognition .

Module8: Sending SMS and Email to Person who is not wearing Mask.

3.6.1 Input images

Dataset consists of two classes such as with mask and without a mask.

3.6.2 Image Preprocessing

Preprocessing is a common and basic procedure that is followed throughout this undertaking.
Preprocessing has the goal of reducing unwanted distortions and improving photo facts as a whole
by improving a few key photo points that may be used in conjunction with processing. The phrase
photo scaling is used to describe the resizing of a digital image in in-camera work and digital
photography. The phrase upscaling or decision enhancement is used in video technology to
describe the amplification of digital content. The picture primitives in an exceeding vector image
photo are scaled using geometric transformations with no loss of photo quality. When scaling a
raster portraiture image, a brand-new snapshot of the top or a reduction in the number of pixels is
expected. When the number of pixels is reduced, a visible first-class loss is expected. One of the
two-dimensional instances of pattern charge conversion in connection to digital sign processing is
thirteen scaling of raster graphics, which comprises the conversion of discrete indications from
one charge to another. The conversion of RGB to grey involves a few steps. The gimp photo
software tool has three methods for this. The lightness technique is a method that uses the most
common of the most notable and least exceptional colorings. The frequent method refers to a
technique that uses a simple three-color common. The luminosity approach is another cutting-
edge technology. Because of reason, the novice is heavily weighted throughout this strategy. For
human perception, frequency is used in addition to the weighted average. Gray to black and white
conversion is accomplished entirely through a binary picture consisting of pixels with exactly one
of the two colorations (black and white). Each pixel in a binary picture is recorded as one bit,
either a zero or a one. Black-and-white, monochromatic, and other terms are frequently used to
describe these photographs. In Photoshop, the bitmap mode is equivalent to a binary image. These
pictures are utilized as masks, thresholding, and dithering in digital photo processing. Bi-level
pictures can only

16
be handled by a few input/output devices, such as laser printers, fax machines, and bilevel laptop
computer displays. A bitmap, which is a dense array of bits, will be used to store a picture.

3.6.3 Data Augmentation

A mask detection tool could no longer take an input file, randomly transform it, and then return
every input and altered data. Keras' image records generator takes the input picture graph and
changes it into 14 modified data at random. Data augmentation may also be defined as a set of
strategies for injecting random oscillations and perturbations and generating a new schooling
sample from an existing one. The model's generalizability is increased by the way facts
augmentation is used.

3.6.4 Model Training

For cellular visual recognition, MobileNetV2 improves on MobileNetV1 in areas like as
classification, item identification, and semantic segmentation. MobileNetV2 is provided as part of
the TensorFlowSlim image classification library. MobileNetV2 modules with pre-trained
checkpoints are also available as TFHub modules. Two new architectural elements have been
added to MobileNetV2:

● linear bottlenecks between layers

● fast links between bottlenecks

Convolutional Neural Network

Convolutional Neural Networks are an artificial neural community that is optimized to technique
pixel information for image cognizance and processing (CNN). The Convolutional Neural
Network is the fundamental and building block of the computer's imaginative and prophetic
assignment of image segmentation.

● Convolutional layer: Filters and kernels are used to create a characteristic map that
summarizes the entered image.

● Pooling layer: This layer is used to summarize the existence of capabilities in function map
patches, which aids in function map downsampling.

● Fully connected layer: Every neuron in a single layer is hooked up to every neuron in every

17
other layer.

Fig 3.1:- Testing the Model Evaluation Metrics

Accuracy
The degree of general correctly identified samples out of all the samples is known as accuracy. It's
characterized as follows:
TP+TN/TP+FP+FN+TN = Accuracy
Where True effectiveness (TP) indicates that something has been successfully acknowledged. False
effective (FP) = erroneously identified True negative (TN) indicates that something has been
successfully rejected. False-negative (FN) indicates that something was rejected mistakenly.

Precision
The amount of effective magnificence forecasts that obviously belong to the effective magnificence
is determined using a precision approach.
Precision = TP/TP+FP
Recall
The number of effective magnificence predictions built from all effective samples within the dataset
is determined using the recall technique.
Recall = TP/TP+FN F1-Score F1- Score is the common suggest of Precision and Recall F1 Score
= 2*(Recall * Precision) / (Recall + Precision)

18
Macro Average
The approach is quite simple. Simply adopt the common sense of precision and maintain the
device with distinct setups in mind. Then Macro-common may be honestly the common
recommendation of Macro-common precision and macro-common remember.

Weighted Average
The F1 Scores are generated for each label, and then their not unusual position is weighted using
a guide - that is, the diversity of real situations for each label. It has the potential to produce an F1
Score that is no longer amid precision and keep in mind.

3.6.6 Image Segmentation using Mask CNN

The Mask R-CNN is a CNN for image segmentation and event segmentation that is built on top of
a quicker R-CNN that can come across objects and limits. Instance segmentation or instance
recognition is the process of detecting all objects in a photograph and segmenting each event. It
emerges as a result of the harsh realities of object detection, localization, and classification
penalties. During this method of segmentation, a clear separation between each item classified as
a similar circumstance is noted. Everyone is treated as a single entity during the event
segmentation process. It's also known as foreground segmentation because it works on the
picture's subject matters rather than the background-CNN and can produce two outputs for each
object, a class label, and a bounding field offset, whereas Mask R-CNN can produce three
outputs, including article masks in addition to the class label and bounding container offset. The
more masks output is exclusive from the opposite two outputs, implying that the finer the spatial
design of an item, the more masks output is required. Mask R-CNN is a faster version of R-CNN
that includes an output for object masks in addition to current outputs such as classification labels
and bounding boxes.

3.6.7 Implementing the Model in OpenCV

Finally, the model is constructed with the help of a webcam, which reads the video frame by
frame and resizes it as needed. Then, in connection with the accuracy in %, the preprocessing
feature is known as achieving the final result of human people wearing masks and now not
wearing a mask.

19
3.6.8 Implementing the Face Recognition
The method of detecting a human face using technology is known as facial reputation. Biometrics
is used in a facial reputation device to map face functions from a picture or video. To find a
match, it compares the information to a database of known faces.
Step 1: Face detection

Both by myself and in a crowd, the digital digicam recognizes and locates a snapshot of a face.
The image might also show the character searching ahead of time or in profile.

Step 2: Face analysis

Following that, a snapshot of the face is taken and examined. Because it may be simpler to
integrate a 2D image with public photographs or those in a database, most facial interest period is
based on 2D as a potential rather than 3D images. The gap between your eyes, the depth of your
eye sockets, the space from the forehead to the chin, the form of your cheekbones, and the outline
of your lips, ears, and chin are all crucial factors that the program interprets. The objective is to
learn the facial landmarks, which are perhaps the most important aspect of facial recognition.

Step 3: Converting the image to data

Face capture mode converts analog facts (a face) into a collection of digital data (data) depending
on the person's facial traits. The appraisal of your face has evolved into a mathematical formula.
A face print is a name for the digital code. Every character has their own faceprint in the same
manner as thumbprints are unique.

Step 4: Finding a match

After that, your faceprint is compared to a database of other recognized faces. Any photo tagged
with a person's name on Facebook becomes a part of Facebook's database, which may be used for
face recognition as well. Self-discipline is achieved when your faceprint matches an image in a
facial interest database.

The face center of attention is the most natural of all the biometric measures. This makes intuitive
sense because we normally recognize ourselves and others by gazing out at faces rather than
thumbprints and irises.

20
 Fig 3.2: - Flow chart for implementation of facial recognition.

3.6.9 Sending SMS and Email to Person who is not wearing Mask
We're employing a face recognition system to identify folks who aren't wearing masks, which will
aid us in locating their information. We will send a preventive message in the form of SMS and
email to such persons in order to reward them, advising them to wear masks in public. We utilized
Twilio for delivering SMS, which provides programmable communication tools for sending and
receiving text messages via its web service APIs, and the smtplib module for email, which simply
establishes an SMTP session for the client object and can then be used to send mail to another
computer.

21
 Fig 3.3:- SMTP protocol

3.7 General Frame of connection of the above modules

22
 Chapter 4
Experimental
Results

1. Testing the Model

Fig 4.1: - Testing the Model

23
2. Detecting Face Mask

Fig 4.2: - Detecting people wearing masks

24
3. Sending Messages via SMS/Email

Fig 4.3:- Sending text messages to people not wearing masks

25
Fig 4.4:- Email template sent to people

26
 Chapter 5
Conclusions

5.1 Conclusions

In this mask detection, we used the MobilenetV2 set of rules to correctly distinguish persons
wearing masks and those without masks, as well as send an email to those involved. Its overall
performance in photographs is generally accurate, and our detecting effects were also rather
accurate. This detection will be utilized for video flow or digital digicam-fed inputs, as well. This
can be used in places of work and establishments by way of training the database with personnel
photos or students' photos and by way of face reputation, and the character is diagnosed by way of
a mobile number that is unique for each person, and other information about the individual is
obtained from the database, and it will be simple to tell that specific character or beneficial for
taking any actions concerned. The suggested version can be improved by including other aspects
such as the number of persons and the social distance between individuals.

5.2 Future Work

Facial recognition with a mask is becoming more and more important over the past year thanks to
the spread of the COVID-19 virus. All through this project, used the MobilenetV2 algorithm and
different deep gaining knowledge of methods to spot humans no longer sporting a mask. This
situation has been tested using a webcam and entering a database, In the future, this function is
often used in conjunction with various AI methodologies and may be performed by units such as
Raspberry Pi, Autonomous drone structures, etc., to improve efficiency and reduce vision time
taken to ensure that people no longer play the mask. and in addition we prefer to combine fines
for those who do not wear the mask properly and measure social distance.
This project will be very useful and can be used in many places like hospitals, airports, schools
offices, shops,, colleges, supermarkets, theaters, temples, houses, etc., and may also be used to
manage free Covid events.

27
 References

[1] Xinbei Jiang, Tianshan Gao, Zichen Zhu, and Yukang Zhao., Real-Time Face Mask Detection
Method Based on YOLOv3, Electronics, 7, pp.130-147, 2021.
[2] Samuel Ady Sanjaya and Suryo Adi Rakhmawan., Face Mask Detection Using MobileNetV2,
International Journal of Engineering and Advanced Technology, 4, pp.2249-8958, 2021.
[3] G. Jignesh Chowdary, Narinder Singh Punn, Sanjay Kumar Sonbhadra, and Sonali Agarwal,
Face Mask Detection using Transfer Learning of InceptionV3, IEEE Access, 20, pp. 456-665,
2021.
[4] Shilpa Sethi, Mamtha Kathuria and Trillok Kaushik., Face mask detection using deep learning,
Multimedia Tools and Application, 8, pp.42-72, 2020.
[5] Chunli Qin, Demin Yao, Yonghong Shi and Zhijian Song, “Computer-aided detection in chest
radiography based on artificial intelligence: a survey,” BioMedical Engineering Online, vol. 17,
pp. 1-23, 2018.
[6] Riya Chiragkumar Shah and Rutva Jignesh Shah., Detection of Face Mask using Convolutional
Neural Network, Mobile Information System, 43, pp.382-487, 2019.
[7] Safa Teboulbi, Seifeddine Messaoud, Mohamed Ali Hajjaji, and Abdellatif Mtibaa., Real-Time
Implementation of AI-Based Face Mask Detection and Social Distancing Measuring System for
COVID-19 Prevention, Scientific Programming, 32, pp.167-254, 2021.
[8] Xueping Su, Meng Gao, Jie Ren, Yunhong Li, Mian Dong, and Xi Liu., Face mask detection
and classification through deep transfer learning, Multimedia Tools and Applications, 53,
pp.11042- 11772, 2021.
[9] Mohamed Almghraby and Abdelrady Okasha Elnady., Face Mask Detection in Real-Time using
MobileNetv2, International Journal Of Engineering and Advanced Technology, 6, pp.49-89, 2021
[10] Chhaya Gupta and Nasib Singh Gill., Corona mask: A Face Mask Detector for Real-Time Data,
International Journal of Advanced Trends in Computer Science and Engineering, 9, pp.2278-
3091, 2021.
[11] Face Mask Dataset – Kaggle Repository.
[12] Source Code Of Project – https://github.com/tushar007giri/Detecting-Mask-over-faces-and-
alerting

28
 Annexure

Source Code of Project –

https://github.com/tushar007giri/Detecting-Mask-over-faces-and-alerting

Annexure A
Train mask detector

29
30
31
Annexure B

Indentifying and matching the details of the persons who have not worn masks.

32
33
34