2. Abstract
Introduction
Existing System
Proposed System
Literature Servey
Data Set
Sequential Model
VGG16 Model
MobileNet V2 Model
Loss and Accuracy Graphs of the Model
Resulting Outputs
References
3. • In the recent years, the Corona viruses disease(COVID-19) which is a big family
of different viruses have become very common, spreadable, contagious and
dangerous to the entire world of human kind. It transmits mostly through nose
and mouth,If a infected person sneeze or cough which leaves droplets of the
virus on different surface which is then inhaled by other person he also catches
the infection too. So it has become very crucial to protect ourselves and the
people around us from this situation. We need to take precautions such as
maintaining social distancing, washing hands every two hours, using sanitizer,
and the most importantly wearing a mask. The usage of wearing mask by public
has become very common everywhere in the entire world. From that like India
there are other countries with vast population in small area are most affected
and in devastating condition to the present scenario of pandemic. This paper
proposes a method to detect the face which is masked or not masked, because
some work places with a lot of people coming to work or to study like schools or
colleges. Here we have used the pretrained models such as the MobileNetV2,
Sequential model, and VGG-16 are used in our context. The model is trained on
a real world data set and tested with live video streaming. Further the accuracy
of the model with different hyper parameters and multiple people at different
distance and location of the frame is done.
4. • Technologies in fields like Machine Learning,
Deep Learning and Artificial Intelligence have
made our lives easier and provide solutions to
several complex problems in various areas. This
model helps in reduce the man power and will
recognise the Masked and Non masked people
through live video Stream where there is overly
crowded places like Educational Institutions,
Shopping Malls, Companies, etc.
5. • In this system if a person wears face mask
Improperly by not covering nose even
though it will get the best output results.
This paper reproduced the training and
testing of the most used classical machine
learning model techniques with opencv,
tensor flow and keras.
6. • We propose a two-stage CNN architecture, where the
first stage detects human faces, while the second stage
uses a lightweight image classifier to classify the faces
detected in the first stage as either ‘Mask’ or ‘No Mask’
faces and draws bounding boxes around them along
with the detected class name.By using imutils we can
integrate the live video cam to detect the with or
without masks persons.
7. Data Set:-
• The Training Data Set that is used in this is the
combination of many images of having with and
without masks.These pictures were taken from
different assets like Kaggle and RMFD datasets. This
data set consists of 2155 Images of data in with mask
category and 2023 Images of data in without mask
category with different Angles.
8. Deep Learning Models That I Have Used
To Find The Accuracy And Loss
Percentage:
• 1.
9. Sequential Model:-
• The Sequential CNN model consists of
layers such as Conv2D, MaXPooling2D,
Flatten, Dropout and Dense. To get the
probability of each class SoftMax function is
used in the Dense layer.
• The Sequential CNN model is trained with
20 epochs which gives us training accuracy
of 0.9899 with loss of 0.0346 beyond which
the accuracy changes due to increased
training time and based upon parameter
values.
CNN Architecture :-
10. VGG-16 Model :-
• Before training with VGG16 model we used
ImageDataGenerator for data augmentation. Later
VGG16 CNN is trained which has 17 convolution layers
and 5 Max Pooling layers.
• The model is trained upon the training dataset and gives
accuracy of 0.9427 with loss of 0.1436.
VGG 16 Architecture :-
11. MobileNet V2 Model :-
• MobileNetV2 architecture has two blocks
of stride 1 and 2 for residual block and
downsizing
• Each block consists of 3 layers such as
1x1 convolution with ReLU6, depth wise
convolution and 1x1 convolution without
non-linearity.
• Model is trained for 20 epochs and gives
accuracy of 0.9922 with loss of 0.0282 for
the training set. Fig. 5 represents the
MobileNetV2 architecture.
MobileNetV2 architecture :-
19. Experimental Results :-
Wearing Mask
Result %
No MaskResult % Half maskResult %
MobileNet Mask : 99.80% No Mask: 100% No Mask: 97.36%
CNN Mask: 97.55% No Mask: 90% Mask : 88.14%
Vgg 16 Mask: 94.26% No Mask: 100% No Mask : 94.62%
20. Future Scope:-
• The present model proposed gives great accuracy for single face
and multiple faces with and without mask also it gives quite good
accuracy. It works easily on any mobile device just by switching on
the video stream, with no external hardware requirement. Further I
will work for improving the accuracy for multiple face mask
detection, to classify the faces into three categories that is, With
mask, without mask, Improper mask instead of just the two with and
without mask class
21. REFERENCES:-
[1] A. G. Howard, M. Zhu, B. Chen et al., “Mobilenets: efficient
convolutional neural networks for mobile vision applications,” 2017,
https://arxiv.org/abs/1704.04861.
[2] Wei Wang, Yutao Li, Ting Zou, Xin Wang, Jieyu You, Yanhong
Luo, "A Novel Image Classification Approach via Dense-MobileNet
Models", Mobile Information Systems, vol. 2020,ArticleID 7602384,
8 pages, 2020. https://doi.org/10.11 55/2020/7602384
[3] I. B. Venkateswarlu, J. Kakarla and S. Prakash, "Face mask
detection using MobileNet and Global Pooling Block," 4 2020 IEEE
4th Conference on Information & Communication Technology
(CICT), 2020, pp. 1-5, doi: 10.1109/CICT51604.2020.9312083.
22. [4] M. S. Ejaz and M. R. Islam, "Masked Face Recognition Using
Convolutional Neural Network," 2019 International Conference on
Sustainable Technologies for Industry 4.0 (STI), 2019, pp. 1-6, doi:
10.1109/STI47673.2019.9068044
[5] Changjin Li, Jian Cao, and Xing Zhang. 2020. Robust Deep
Learning Method to Detect Face Masks. In Proceedings of the 2nd
International Conference on Artificial Intelligence and Advanced
Manufacture (AIAM2020). Association for Computing Machinery,
New York, NY, USA, 74–77.
DOI:https://doi.org/10.1145/3421766.3421768
[6] K. Simonyan and A. Zisserman, “Very deep convolutional
networks forlarge-scale image recognition,” CoRR, vol.
abs/1409.1556, 2014.