AI-Based Low-Cost Real-Time Face Mask Detection and Health Status

Monitoring System for COVID-19 Prevention

CHOON EN YOU, WAI LEONG PANG, KAH YOONG CHAN

Faculty of Engineering,

Multimedia University,

63100 Cyberjaya,

MALAYSIA

Abstract: - The outbreak of COVID-19 had brought a great challenge for the World Health Organization

(WHO) in preventing the spreading of SARS-CoV-2. The Ministry of Health (MOH) of Malaysia introduced

the MySejahtera mobile application for health monitoring and contact tracing. Wearing a face mask in public

areas had been made compulsory by the Government. The overhead cost incurred in hiring the extra manpower

to ensure all the visitors wear a face mask, check-in through MySejahtera and the status in MySejahtera is

healthy before entering a premise. A low-cost solution is urgently needed to reduce the heavy overhead cost.

An AI-Based Low-Cost Real-Time Face Mask Detection and Health Status Monitoring System (AFMHS) is

proposed to perform real-time detection for the face mask and MySejahtera Check-In tickets by using artificial

intelligence. MobileNetV2 was used for the detection and recognition of face and face masks. YOLOv3 was

used for the detection of the region of interest for the MySejahtera Check-In ticket to locate the health and

vaccination status of the visitor. Optical character recognition (OCR) is a technique that is used to detect the

text captured in an image and encode the recognized text. OCR is implemented to recognize the text extracted

from the ticket. Tesseract is used as the OCR engine in AFMHS. Raspberry-Pi-4B (Raspberry Pi Generation 4

Model B) with 4 GB RAM is used as the processing unit of AFMHS. The total cost of the AFMHS is only

USD220. Extensive experimental tests were carried out to evaluate the performance of AFMHS. The optimum

operation setup conditions are proposed to achieve 100% accuracy. The optimum operating distance for the

face mask detector and MySejahtera Check-In ticket detector are 1.5m and 15cm respectively.

Key-Words: - AI, COVID-19, Face mask detection, Machine Learning

Received: June 13, 2021. Revised: September 8, 2022. Accepted: October 8, 2022. Published: November 4, 2022.

1 Introduction

On 11 March 2020, the WHO declared a pandemic

that had been caused by the coronavirus disease

namely COVID-19. COVID-19 caused the severe

acute respiratory syndrome. The pandemic forced

the Government of Malaysia to take action to

prevent the spreading of the COVID-19 disease. The

Government of Malaysia had implemented several

standard operation procedures (SOP) to fight the

pandemic, i.e. wearing a face mask in public, and

introduced the MySejahtera mobile application for

contact tracing and health status. MySejahtera

assists in suppressing the COVID-19 outbreak by

helping users in monitoring their health conditions.

Users could report their health status when COVID-

19 symptoms were detected. The detailed

information of the users that are infected by

COVID-19 is recorded and the COVID-19 risk

status is updated accordingly in the application.

MySejahtera Check-In function is important for

all types of public premises such as restaurants,

shops, companies, schools and construction sites.

All public premise owners need to display the QR

code generated by MySejahtera. Visitors are

required to use the MySejahtera QR Code Scanner

to scan the QR code displayed before entering any

public premises. Upon scanning, a ticket will be

generated whereby information such as the name of

public premises, name of the individual who entered

the public premise, dates and phone number will be

listed. The health risk and vaccination status of the

visitor would also be shown. Only the visitor with

the status of low risk (no symptoms) and fully

vaccinated could enter the premises. This is another

mandatory rule set by MOH.

The flow of people that would pass in and out of

public premises such as schools, shopping malls,

shops and restaurants is high. It would increase the

difficulty of the premise owner to ensure that the

people that enter the premises are healthy, and

checked in through MySejahtera with a healthy and

vaccinated status. It is a huge challenge to guarantee

that the visitors would wear the face mask correctly

and have scanned the QR code before entering the

WSEAS TRANSACTIONS on INFORMATION SCIENCE and APPLICATIONS

DOI: 10.37394/23209.2022.19.26

Choon En You,

Wai Leong Pang, Kah Yoong Chan

E-ISSN: 2224-3402

256

Volume 19, 2022

premises. The premises’ owners hired security

guards to ensure that all the visitors are wearing face

masks, with normal body temperature and also

check the health and vaccination status in the

MySejahtera of the visitors at the entrance of the

premise. Only visitors with normal body

temperature and healthy MjSejahtera status are

allowed to enter the premises. It could be a huge

financial burden to the premise owner as the extra

cost is incurred. From the perspective of an

economical view, this effect is greatly obvious if

several security guards are being hired to monitor

the premise. These have inspired us to propose an

AI-Based Low-Cost Real-Time Face Mask

Detection and Health Status Monitoring System

(AFMHS).

The project objectives are to develop a real-time

alert system for the face mask detector and

MySejahtera Check-In ticket detector using artificial

intelligence. The MobileNetV2 algorithm is used in

the face mask detector [3]. Optical character

recognition (OCR) is incorporated with the

YOLOv3 object detection algorithm to detect the

specific targeted character from the MySejahtera

Check-In ticket [4]. The AFMHS allows the visitor

who wears a face mask and has a record of low risk

with fully vaccinated on the MySejahtera Check-In

ticket to enter the premises.

2 Literature Review

In the year 2021, the researcher Fushuai Wang et al

did a research project on face recognition with

MobileNetV2. The processing unit proposed is on

Raspberry Pi 4B, [1]. The author mentioned that

MobileNetV2 could provide a balance between the

accuracy and the network parameter, which allows it

to be suitable for usage on mobile devices. The face

images were recognized in a total of 5 categories.

100 face images are used in the training set which is

500 images in total. 50 face images in the test set,

which means 250 images in total, [1].

In the year 2021, the researcher Ikram ben Abdel

Ouahab et al developed a real-time face mask

detector with MobilNetV2 and Raspberry Pi, [2].

The face mask model was trained, tested and

evaluated through Google Colab with a huge

database available online on GitHub. The dataset

consists of a total of 3835 images with 1915 images

with masks and 1918 images without mask. The

selected database consists of people wearing masks

with different poses and positions, [2]. The

approach of this project work was more towards the

ability to perform in real-time whereby speed is a

crucial factor. A smooth motion and appearance of

the video stream were considered. When the real-

time video stream is analysed by the detection

model, the frame per second (FPS) value will

decrease. The decrease in FPS will be even more

obvious in devices with lower processing power.

From this research work, the lower processing

power device such as Raspberry-Pi-4B could cater

for the performance requirement. MobileNetV2

could also be deployed due to the nature of its

lightweight model.

In the year 2021, the researcher Samuel Ady

Sanjaya et al completed research work on face mask

detection by using MobileNetV2, [3]. In this

research paper, the author mentioned that face mask

recognition was being implemented with the image

classification method through MobileNetV2. The

proposed face mask detection model was referring

to the datasets taken from the Kaggle dataset and the

Real-World Masked Face dataset. These datasets

consist of 5 thousand masked faces and 90 thousand

normal faces photos.

Extra work is carried out in the year 2022 to

perform face mask detection. The enhanced Yolo

algorithm was proposed to detect the face mask, [9].

Experimental work was carried out to evaluate the

performance of the algorithm proposed and this

shows the importance of the solution in face mask

detection. Convolutional Neural Network was used

to monitor the face mask of the workers on the

construction site to ensure the workers’ safety, [10].

The solution proposed also monitors the physical

distance between the workers. The performance of

various approaches in face mask detection was

reviewed in [11]. However, no related work is

reported in monitoring the health status of

MySejahtera. An effective solution is urgently

needed to monitor the health status in MySejahtera

of the people before they enter the premises.

In the year 2021, the researcher R Shashidhar et

al produced a project regarding the detection and

recognition of vehicle number plates with the

method of OCR through YOLOv3, [4]. Since there

are different background colours and types of

license plate, the YOLOv3 model was trained to

localize the vehicle number plate. It is particularly

mentioned that YOLO had been used to find the

region of interest which is the part of the vehicle

number plate only [4]. A dataset that consisted of

6439 images of different alphabet-numerical

characters was created. The result showed that an

accuracy of 91.5% was obtained for this vehicle

number plate detection model ,[4].

In the year 2020, the researcher Chinmaya

Kumar Sahu et al did research work on a

comparative analysis of the deep learning approach

WSEAS TRANSACTIONS on INFORMATION SCIENCE and APPLICATIONS

DOI: 10.37394/23209.2022.19.26

Choon En You,

Wai Leong Pang, Kah Yoong Chan

E-ISSN: 2224-3402

257

Volume 19, 2022

regarding YOLOv3 for real-time recognition of

vehicle number plates, [5]. 2500 images of cars that

consist of foreign number plates and Indian license

plate numbers had been prepared. The images will

be labelled with a desktop application namely

Labeling. An Extensible Markup Language (XML)

file is created for each labelled image. The region of

interest will be labelled with a bounding box. Upon

labelled, the coordinate and the object class name of

the region of interest that contained the target

outcome are stored in the xml text file. The

coordinate stored in the xml file is compatible with

the Darknet Config. Darknet Config data

architecture is part of the YOLOv3 model. After the

annotation and labelling of the dataset, the YOLOv3

model was trained for the localization of vehicle

number plate detection.

In the year 2022, the researcher Srividya

Subramanian et al proposed a text extraction model

with YOLOV3, [6]. YOLOv3 has been

implemented to localise and classify the characters

from images. Newspapers and books in the form of

digital were collected and used as the dataset. There

were a total of 80 distinct characters being used. An

annotation process had been carried out to draw the

bounding box over the characters. An annotation file

that contained the coordinate of the location of the

character had been created.

A comparative review had been made on 4

different OCR engines. The researcher Ahmad P.

Tafti et al had done a comparative review to

evaluate the OCR engine between Google Docs

OCR, Tesseract, ABBYY FineReader as well as

Transym, [7]. From this research work,

experimental evaluation in the form of qualitative

and quantitative approaches had been made using

these OCR engines and services. In order to perform

the experimental comparison, 1227 images from 15

different categories had been used, [7]. The main

focus will be on the category of machine-written

characters, machine-written digits, noisy images and

blurred images. Tesseract is a good choice since the

service provided is free of charge.

3 Design of the AFMHS

Fig. 1 shows the two main components of the

AFMHS, i.e. the face mask detector and the

MySejahtera Check-In ticket detector. The object

detection algorithm is modelled with the python

programming language. OpenCV is a Python library

for image processing to perform face recognition

and object detection. OpenCV has been installed in

Python to perform face detection and image

processing. Imutils is used for image processing and

the functions of imutils include rotation and resizing

of images. The images will be preprocessed by

Imutils before being accessed by the OpenCV.

Tensorflow is used for machine learning

applications.

Fig. 1: Two main components of the AFMHS

3.1 Face Mask Detector

The dataset used consists of images of one

individual and more than one individual. Different

conditions of content are covered in the dataset.

Some of the images are slightly blurred and the

pixel is not as high. The pictures are taken in

various positions and different poses. The images of

the users wearing the different types of face masks

are collected as well. There are various types and

colours of masks collected in the dataset including

normal medical face masks and N-95 face masks. A

dataset that is rich in a variant of information that is

reasonable and related is preferred.

Scikit-learn is used for image classification and

regression. The dataset consists of two groups of

images. One group of images for faces with masks

is saved in a file namely with_mask. Another group

of images for faces without masks is saved in a file

namely without_mask. The number of images for

the file “with_mask” is 504 images and 430 images

AFMHS

1. Face

Mask

Detector

MySejahtera

Check-In

Ticket

Detector

Face mask

detected

No mask

detected

Green LED =

Buzzer = Off

"Low" and

"Fully"

detected

"High" or

"Not"

detected

Green LED =

Off

Buzzer = Beep

Green LED =

Buzzer = Off

Green LED =

Off

Buzzer = Beep

AFMHS

1. Face

Mask

Detector

MySejahtera

Check-In

Ticket

Detector

Face mask

detected

No mask

detected

Green LED =

Buzzer = Off

"Low" and

"Fully"

detected

"High" or

"Not"

detected

Green LED =

Off

Buzzer = Beep

Green LED =

Buzzer = Off

Green LED =

Off

Buzzer = Beep

AFMHS

1. Face

Mask

Detector

MySejahtera

Check-In

Ticket

Detector

Face mask

detected

No mask

detected

Green LED =

Buzzer = Off

"Low" and

"Fully"

detected

"High" or

"Not"

detected

Green LED =

Off

Buzzer = Beep

Green LED =

Buzzer = Off

Green LED =

Off

Buzzer = Beep

WSEAS TRANSACTIONS on INFORMATION SCIENCE and APPLICATIONS

DOI: 10.37394/23209.2022.19.26

Choon En You,

Wai Leong Pang, Kah Yoong Chan

E-ISSN: 2224-3402

258

Volume 19, 2022

for “without_mask” which made out a total of 934

images.

The dataset is used by the MobileNetV2 neural

network for the training process. All the images had

been resized to 224 x 224 pixels. Next, another pre-

processing step is the conversion of the images into

the NumPy array format. All the images of the

dataset had been scaled to the range of -1 to 1. The

preprocessing step is to perform one-hot encoding

on the labels. 80% of the images from the dataset

were split into training and another 20% for testing.

The MobileNet model is loaded together with pre-

trained ImageNet weights.

The FaceNet model is implemented for face

detection. The installed and imported packages are

used to grab the dimension of the frame. A blob is

created from the frame. Afterwards, the FaceNet

model will be deployed for face detection when the

created blob is being passed through the network.

The confidence in the detection is extracted.

Detections with confidence levels lower than 0.5 are

filtered out and excluded. Next, the x and y

coordinates of the bounding box for the detected

object are calculated. The frames from the real-time

streaming video are looped over.

Detection of the faces in the frame is carried out.

The face in the bounding box is processed and

determined whether the face is wearing a face mask

or not. The predictions for both labels are compared

and displayed on top of the bounding box. The label

is either “Mask Detected” on top of the green colour

bounding box or “No Mask Detected” on top of the

red colour bounding box.

3.2 MySejahtera Check-In Ticket Detector

YOLOv3 will be used as the object detection model

with the backbone of Darknet-53. A dataset that

consists of the images taken from the MySejahtera

Application. A sample of the MySejahtera Check-In

Ticket is shown in Fig. 2. A MySejahtera Check-In

ticket consists of the information regarding the

name of the premise, name, phone number, date,

time, risk status and vaccination status. All this

information is collected. The main focus is put on

the section on risk status and vaccination status. A

total of 210 images had been included in the dataset.

Images of various conditions regarding the risk and

vaccination status of the MySejahtera Check-In

ticket had been collected.

Fig. 2: Annotation of the MySejahtera Check-In

Ticket image by labelImg

There are unnecessary texts from the ticket,

hence it is important to perform annotations to set

the region of interest. In order to perform

annotation, labelImg (a graphical image annotation

tool) is used to draw the bounding box for

identifying the area of interest as shown in Fig. 2.

The regions of interest are the risk status and

vaccination status. However, other information such

as name, date, time, location, and phone number are

annotated as well for future research. There are a

total of 7 bounding boxes drawn for each image. An

annotation text file is generated and it consists of the

coordinates of the bounding boxes. The format of

the position of the bounding box is as follows,

<class name> <x coordinate of centre> <y

coordinate of centre> <width> <height>.

Google Colab is used for the training process. All

the datasets including the images and all the

annotation files are uploaded to Google Drive and

Address

Name

Contact No.

Date

Time

Risk status

Vacination status

WSEAS TRANSACTIONS on INFORMATION SCIENCE and APPLICATIONS

DOI: 10.37394/23209.2022.19.26

Choon En You,

Wai Leong Pang, Kah Yoong Chan

E-ISSN: 2224-3402

259

Volume 19, 2022

directed to Google Colab. 80% of data are used for

training and another 20% for testing. YOLOv3 with

backbone Darkent-53 is used for the training. Each

training weight is saved after each epoch. The best

weight that has the lowest loss is updated and saved.

Tesseract OCR is used as the character

recognition engine. Gray-scaling of the images

before OCR detection is implemented. The purpose

of this preprocessing step is to make sure that less

information is needed for each pixel. It speeds up

the recognition of the character. The MySejahtera

Check-In ticket detector mainly searches for the

word “Fully”, “Low”, “High” and “Not” from the

text file.

3.3 The Operation of the AFMHS

A microcontroller is required to deploy the AFMHS

and a camera module is needed to capture the video.

AFMHS is deployed in a microcontroller,

Raspberry-Pi-4B (Raspberry Pi Generation 4 Model

B) with 4 GB random access memory. AFMHS

used the camera of a Samsung S9 smartphone to

capture the video through the DroidCam software,

[8]. Droidcam is an application that would provide

the service of using a smartphone as a webcam. The

Samsung S9 smartphone consists of an 8MP front

camera and a 12MP rear camera which can capture

video with good resolution for AFMHS. An alert

system is proposed in AFMHS. This system consists

of a green LED and a buzzer. The green LED acts as

an indicator for the user that fulfills the requirement

(with face mask, healthy and vaccinated status) to

pass through while the buzzer acts as an alarm

system to alert those who are not fulfilling the

requirement.

Fig. 3 shows the system operation flowchart of

the AFMHS proposed. AFMHS consists of two

functions, i.e. 1) Face mask detector and 2)

MySejahtera Check-In Ticket Detector. The face

mask detector captures the video through the

camera. The green LED is on once the visitor is

wearing the face mask and the system will show

“Mask Detected” on the screen. However, AFMHS

will show “No Mask Detected”, the green LED is

turned off and the buzzer is beeping when the visitor

is not wearing a face mask detected.

The second function of AFMHS is MySejahtera

Check-In Ticket Detector. AFMHS analyses the risk

and vaccination status shown in MySejahtera

Check-In Ticket (Fig. 2). Visitors with “Low” risk

and “Fully” vaccinated status are granted to enter

the premises. The green LED is turned on to

indicate that the visitor is healthy, fully vaccinated

and safe to enter the premises. The buzzer will beep

once “High” risk or “Not” vaccinated are detected.

AFMHS alerts the authority that the visitor with

either at high risk or not vaccinated is detected and

is not allowed to enter the premises.

Fig. 3: The operation flowchart of AFMHS

4 Results and Discussion

Extensive experimental tests are carried out to

determine the accuracy of the face mask detector

with the distance between the individual and camera

START

- Turn on the Face Mask Detector

- Turn on the camera

Is face mask

detected?

- Buzzer beeps

- Screen shows

“No Mask

Detected”

- Turn off the

Green LED

- Turn off the Buzzer

- Screen shows “Mask Detected”

- Turn on the Green LED

Is “Low” & “Fully”

detected?

Is “High” or

“Not” detected?

- Buzzer beeps

- Turn off the Green

LED

- Turn off the Buzzer

- Turn on the Green

LED for 10 seconds

- Turn off the Green LED

- Turn off the Check-In Detector

END

Yes

- Turn off the Face Mask Detector

- Turn on the Check-In Detector

- Turn off the Green LED

Yes

WSEAS TRANSACTIONS on INFORMATION SCIENCE and APPLICATIONS

DOI: 10.37394/23209.2022.19.26

Choon En You,

Wai Leong Pang, Kah Yoong Chan

E-ISSN: 2224-3402

260

Volume 19, 2022

as the fixed variables and the light intensity of the

environment as manipulating variables. Fixed

distances of 1m, 1.5m, 2m, 2.5m, and 3m are used.

The light intensities of 40~60 Lux, 90~120 Lux,

200~220 Lux, and 300~320 Lux are used. Three

different types of face masks are used which are the

black mask, the green mask, and the N95 white

mask. AFMHS is tested for the accuracy of

detecting no mask as well.

Table 1. Accuracy with different light intensities

Distance

(m)

Light

intensity

(Lux)

Accuracy

Black

Green

White

(N95)

mask

1.0

300~320

100%

200~220

100%

90~120

100%

40~60

100%

1.5

300~320

100%

200~220

100%

90~120

100%

40~60

100%

2.0

300~320

60%

100%

200~220

50%

100%

90~120

50%

100%

40~60

40%

100%

2.5

300~320

60%

100%

200~220

40%

80%

100%

90~120

40%

60%

100%

40~60

10%

60%

80%

3.0

300~320

20%

40%

200~220

20%

90~120

20%

40~60

20%

Table 2. Accuracy of the MySejahtera Check-In

ticket detector with fixed light intensity (60 Lux)

and anti-glare hood

Distance (cm)

Accuracy

10/10 = 100%

1/10 = 10%

0/10 = 0%

In order to test the functionality and limit of the

prototype, several parameters and conditions had

been evaluated. For each parameter, the experiment

is repeated 10 times. The measured output results

are recorded. The output is correct when it is the

same as the expected output. The output is wrong

when it is different from the expected output or no

output had been produced. For each parameter, the

accuracy rate is based on the correctness of the

output result of AFMHS measured.

As shown in Table 1, the longer the distance, the

lower the accuracy. The distance of 1m and 1.5m is

the optimum distance value as the performance in

terms of accuracy for all categories is high. The

lower the light intensity, the lower the accuracy. The

optimum light intensity is 300~320 Lux. Moreover,

the face mask detector had been tested for the

incorrect way of wearing the mask with a fixed light

intensity of 300~320 Lux and the distance is fixed at

1.5m. The accuracies across all categories of masks

are between 80% to 90%.

Fig. 4 shows the results of the face mask

detector. Fig. 4(a) shows the figure when the face

mask is detected. A green colour bounding box with

“Mask Detected” on top of the box is shown in the

figure when the face mask is detected. Fig. 4(b)

shows the figure when no mask is detected. A red

colour bounding box with “No Mask Detected” on

top of the box is shown in the figure when no mask

is detected.

The glare on the handphone screen affected the

accuracy level of the detection. The glare on the

screen came from the reflection of the light source.

The mitigation is to build an anti-glare camera hood

to place the camera device in order to enhance the

AFMHS performance. The purpose of the anti-glare

camera hood is to limit the glare on the handphone

screen in order to enhance the image quality of the

MySejahtera Check-In Ticket.

Without the anti-glare hood, the light intensity

around the screen of the phone is high. With a fixed

distance of 0.08m between the camera and the

screen of the user’s phone that displays the

MySejahtera Check-In ticket, the light intensities of

150 Lux, 200 Lux, 300 Lux, and 500 Lux were used

and the accuracies were 80%, 80%, 60%, 50% and

30% respectively. When an anti-glare hood was

implemented to avoid the reflection of light and

glare on the phone’s screen, the accuracy of the

detector increased. With the anti-glare hood, the

light intensity around the phone’s screen is reduced

to the range of 20 to 60 Lux. Although the light

intensity is low, the recorded accuracy of the

detector is excellent with an accuracy of 100%.

Furthermore, with an anti-glare hood, the

distance of 8cm, 12cm, 15cm, 20cm and 25cm

between the camera and the user’s phone and the

light intensity of 60 Lux were used to evaluate the

performance of the MySejahtera Check-In ticket

detector. Fig. 5 shows the results of the MySejahtera

Check-In ticket detector. As shown in Table 2, it

was observed that the longer the distance, the lower

the accuracy. The accuracy starts to drop when the

distance is more than 15cm. Hence, the optimum

distance is between 8cm to 15cm. It can be deduced

WSEAS TRANSACTIONS on INFORMATION SCIENCE and APPLICATIONS

DOI: 10.37394/23209.2022.19.26

Choon En You,

Wai Leong Pang, Kah Yoong Chan

E-ISSN: 2224-3402

261

Volume 19, 2022

that in order to achieve high accuracy for the

MySejahtera Check-In ticket detector, the distance

should not be more than 15cm and the light intensity

of the environment should keep at least around

20~60 Lux with the anti-glare hood.

(a) Face mask detected

(b) No mask detected

Fig. 4: Results of face mask detector

Fig. 5: Results of the MySejahtera Check-In ticket

detector

5 Conclusion

The AFMHS proposed is a perfect solution to

ensure the users wearing a face mask and the health

status in MySejahtera are healthy and vaccinated.

No similar work is reported in monitoring the health

and vaccination status in MySejahtera. An effective

and low-cost AFMHS is urgently needed. AFMHS

consists of two major functions, i.e. face mask

detector and MySejahtera Check-In ticket detector.

Extensive experimental works are carried out to

enhance the performance of AFMHS. The anti-glare

hood is proposed to improve the accuracy of the

AFMHS. The optimum conditions for the face mask

detector are fixed light intensity of 300~320 Lux

and the distance is fixed at 1.5m. The optimum

experimental setup of AFMHS is proposed to

enhance its performance. The distance between the

camera and the user’s phone should not be more

than 15cm and the light intensity of the environment

should keep at least around 20~60 Lux for the

MySejahtera Check-In ticket detector. The accuracy

of the AFMHS is 100%. The total cost of the

AFMHS is only USD220. It is a promising solution

that is ready to be deployed and is more economical

compared to the current practice in which extra

manpower is used to ensure the visitors are wearing

face masks and the MySejahtera status is healthy

before entering a premise.

Acknowledgement:

This research was funded by the Internal Research

Fund, Multimedia University, Malaysia.

WSEAS TRANSACTIONS on INFORMATION SCIENCE and APPLICATIONS

DOI: 10.37394/23209.2022.19.26

Choon En You,

Wai Leong Pang, Kah Yoong Chan

E-ISSN: 2224-3402

262

Volume 19, 2022

References:

[1] F. Wang, R. Zheng, P. Li, H. Song, D. Du and J.

Sun, Face recognition on Raspberry Pi based on

MobileNetV2, 2021 International Symposium

on Artificial Intelligence and its Application on

Media (ISAIAM), Xi'an, 2021, pp. 116–120.

[2] O. Abdel, L. Elaachak, M. Bouhorma and Y. A.

Alluhaidan, Real-time Facemask Detector using

Deep Learning and Raspberry Pi, 2021

International Conference on Digital Age

Technological Advances for Sustainable

Development (ICDATA), Marrakech, 2021, pp.

23–30.

[3] S. A. Sanjaya and S. R. Adi, Face Mask

Detection Using MobileNetV2 in The Era of

COVID-19 Pandemic, 2020 International

Conference on Data Analytics for Business and

Industry: Way Towards a Sustainable Economy

(ICDABI), University of Bahrain, 2020, pp. 1–5.

[4] R. Shashidhar, A. S. Manjunath, R. K.

Santhosh, M. Roopa and S. B. Puneeth, Vehicle

Number Plate Detection and Recognition using

YOLO- V3 and OCR Method, 2021 IEEE

International Conference on Mobile Networks

and Wireless Communications, Karnataka,

2021, pp. 1–5.

[5] C. K. Sahu, S. B. Pattnayak, S. Behera and M.

R. Mohanty, A Comparative Analysis of Deep

Learning Approach for Automatic Number Plate

Recognition, 2020 Fourth International

Conference on I-SMAC (IoT in Social, Mobile,

Analytics and Cloud) (I-SMAC), Palladam,

2020, pp. 932–937.

[6] S. Subramanian, V. Kekatpure, G. Raymond, K.

Parab, S. Dugad and A. Shirke, TEYSuR - Text

Extraction with YOLO and Super Resolution,

2022 International Conference for Advancement

in Technology (ICONAT), Maharashtra, 2022,

pp. 1–7.

[7] A. P. Tafti, A. Baghaie, M. Assefi, H. R.

Arabnia, Z. Yu and P. Peissig, OCR as a

Service: An Experimental Evaluation of Google

Docs OCR, Tesseract, ABBYY FineReader, and

Transym, Lecture Notes in Computer Science,

Vol. 10072, 2016, pp. 735–746.

[8] “Dev47Apps: DroidCam - Use your phone as a

webcam!” https://www.dev47apps.com/

(accessed Oct. 28, 2022).

[9] P. Wu, H. Li, N. Zeng and F. Li, FMD-Yolo:

An efficient face mask detection method for

COVID-19 prevention and control in public,

Image and Vision Computing, Vol. 117, 2022,

pp. 1-10.

[10] M. Razavi, H. Alikhani, V. Janfaza, B. Sadeghi

and E. Alikhani, An Automatic System to

Monitor the Physical Distance and Face Mask

Wearing of Construction Workers in COVID-19

Pandemic, SN Computer Science, Vol. 3, 2022,

pp. 1-8.

[11] Vibhuti, N. Jindal, H. Singh and P. Rana, Face

mask detection in COVID-19: a strategic

review, Multimedia Tools and Applications,

Vol. 81, 2022, pp. 40013–40042.

Contribution of Individual Authors to the

Creation of a Scientific Article (Ghostwriting

Policy)

Choon En Yuo carried out the simulation,

experimental test and optimization.

Wai Leong Pang has carried out the

conceptualization of the project and carried out the

formal analysis, validation, supervision, review and

edit of the paper.

Kah Yoong Chan has carried out the validation,

project management, review and edit of the paper.

Creative Commons Attribution License 4.0

(Attribution 4.0 International, CC BY 4.0)

This article is published under the terms of the

Creative Commons Attribution License 4.0

https://creativecommons.org/licenses/by/4.0/deed.en

_US

WSEAS TRANSACTIONS on INFORMATION SCIENCE and APPLICATIONS

DOI: 10.37394/23209.2022.19.26

Choon En You,

Wai Leong Pang, Kah Yoong Chan

E-ISSN: 2224-3402

263

Volume 19, 2022