Professional Documents
Culture Documents
ISSN No:-2456-2165
Abstract:- COVID-19 has affected the world badly. II. RELATED WORK
Studies have demonstrated that wearing a facial
covering is one of the insurances to diminish the danger Sahana Srinivasan et.al [1] provides a comparative
of viral transmission. And many public places as well as study of different face detection and face mask classification
public service providers require customers to use the models. Finally, a video dataset labelling method is
service and place only if they wear mask correctly. proposed along with the labelled video dataset to
Sometime it may not be easy to manually track the compensate for the lack of dataset in the community and is
customer, whether they are wearing the mask or not. used for evaluation of the system. The system performance
That’s why this technology holds the key here. In this is evaluated in terms of accuracy, F1 score as well as the
system, we propose face mask detection using image prediction time, which has to be low for practical
processing which is one of the high-accuracy and applicability. The system performs with an accuracy of
efficient face mask detector. This proposed system is of 91.2% and F1 score of 90.79% on the labelled video dataset
three stages i.e. 1. Image preprocessing 2. Face detection and has an average prediction time of 7.12 seconds for 78
and crop 3. Face mask classifier. Our system is capable frames of a video.
of detecting masked and unmasked faces and can be
integrated with cameras and other hand detecting the In [2] proposed a system where For evaluation of the
trained model, mAP (Mean Average Precision) was
distance between two people whether they are maintain
calculated for both the use cases (Social Distancing & Face
distance or not .
Mask Detection), it works by comparing the ground-truth
Keywords:- Covid_19, Image Processing, Mask, CNN , bounding box vs the detected box and, in the end, returns the
SSD. score. The higher the mAP score would be, the better model
is in the detection of objects. Mean Average Precision was
I. INTRODUCTION calculated for two different thresholds (0.25 % & 0.50 %)
with 101 recall points. Three different classes were created
In Wuhan, China at the end of 2019 Corona Virus was for classification those were Good, Bad & None, for which
detected. From that time, it has been spreading like a wild True Positive & False Positive values were calculated with
fire in a timber area. Millions have been affected and around ROC Curve for better understanding.
have unfortunately passed away as on 30th of December
2020, nearly a time since this contagion came to actuality. In [3] surveys various deep learning networks to
People who have this illness can take up to 2 weeks to cure, develop such detectors. In this survey, the existing object
with the threat of having to suffer fresh medical problems detection models used for surveillance and people detection
caused by it. Kiddies and senior folks individualities have are analyzed. The one-stage and two-stage detectors along
ended up being at the most elevated peril to get the with their applications and performance are outlined in a
infection, which might indeed bring about death. Latterly, it comprehensive manner. Deep Learning models such as
has been concentrated on to contain the infection than to fix AdaBoost, Voila-Jones, variants of CNN including ResNet,
it. The infection spreads through the air, communicated by VGG-16, single-shot detectors MobileNet, and versions of
one existent to another by contact, yet also by talking and YOLO are discussed and compared.
playing. The solicitude was advanced to WHO (World
Health Organization) which recommended that facial In[4] system focuses on a solution to help enforce
coverings and social removing is the response to it, until a proper social distancing and wearing masks in public using
fix is created. Putting a facial covering on can dwindle the YOLO object detection on video footage and images in real
peril of getting tainted by an extraordinary degree, not time. The experimental results shown in the paper infer that
simply to the one wearing it yet also to the others that he the detection of masked faces and human subjects based on
interacts with. Wearing curtains each time we go out is YOLO has stronger robustness and faster detection speed as
commodity we can do with little exertion that can compared to its competitors. Their proposed object detection
adequately save lives, and that's definitively why it's in such model achieved a mean average precision score of 94.75%
a lot of interest now of time. Hence we've proposed a system with an inference speed of 38 FPS on video.
with two modules i.e. Face mask and social distancing
In [5] proposed a system where they have taken one of
the measures used to prevent COVID-19 spread and aimed
to develop a deep learning model to categorize people with
or without a mask at public places such as schools, colleges,
and corporates. Developed algorithm using concepts of deep