Intelligent Mask Detection Using Deep Learning Techniques

Owing to the corona pandemic, the government has insisted on wearing a safety mask and maintaining 6 feet distance to get rid of CoronaVirus. The detection of people with or without masks is a challenge due to the impact of Covid pandemic. There are some models / systems which really reduce the manpower to notify the people. The existing system runs on the model: Yolov3, V G G, for face detection and MobileNetv2 for face recognition, object detection, and semantic segmentation inorder to detect the people with and without masks. The proposed system holds an approach of detecting human’s faces and classifying them into people with and without masks which has been done using image processing and deep learning and our project runs u.3nder a model called Faster RCNN. Moreover, Faster R-CNN is more accurate while other models are faster. Being effective is not important but being efficient is way more important.


Introduction
The arena is struggling with Covid19 pandemic. There are such countless essential sorts of prevention expected to battle against Corona infection. One in every of such most fundamental is Face mask. First and fundamental face cover become now not obligatory for absolutely everyone but alternatively because the day advances researcher and medical doctors have prescribed anybody to put on face veil. currently To recognize if an character is carrying Face mask, we are able to make use of Face masks Detection method. Face masks Detection Platform makes use of artificial network to peer if a man or woman does/would not put on a cowl. The software can be related with any present day or new IP cameras to recognize people with/without a veil. in this weblog we will see several vast a part of face cover discovery for Covid19 cases in addition to for different normal cases. The sample of sporting face covers overtly is ascending because of the Coronavirus Covid pestilence everywhere on the arena. Previously Coronavirus, humans used to put on veils to at ease their wellbeing from air infection. Whilst others are unsure approximately their looks, they hide their feelings from human beings in popular by way of concealing their countenances. Researchers sealed that sporting face covers deals with blockading COVID-19 transmission.

Related Works
The situation report ninety six of global health affiliation (WHO) [1]  infection 2019 (COVID-19)has tainted over 2.7 million individuals and induced more than 180,000 people. Likewise, there are some genuine respiratory illnesses, like extreme severe respiratory disease, which passed off within the previous couple of periods. [2][3][4] found out that the conceptive wide variety of Coronavirus is better contrasted with the SARS. [5] displayed that the cautious face covers should reduce impact on Covid 19. WHO suggests that people need to wear face covers at the outside area, if they have any respiratory symptoms, or they're dealing with some manifestations [6]. Viola Jones locator makes use of Haar encompass with vital photo strategy [7], whilst extraordinary works acquire various element extractors, like histogram of situated tendencies (HOG), scale-invariant detail exchange (SIFT), and so forth [8]. As of late, these object detecting models exhibit extra special execution and overwhelm the development of current object identifiers. One-level identifiers make use of a solitary neural agency to distinguish gadgets, consisting of single shot indicators (SSD) [9] and also you simply look as soon as (YOLO) [10]. YOLO separated the photograph into a few cells and in a while attempted to coordinate the anchor containers to objects for every cellular, but it is limited for little objects [11]. Single-shot Detector (SSD) conducts discovery on a few element courses to identify faces in diverse sizes [12]. To improve precision, [13] proposes RetinaNet with the aid of becoming a member of SSD and FPN engineering. MobileNet makes use of smart convolution to take away highlights to alternate channel numbers, so the computational cost of MobileNet decreases than networks making use of well-known convolutions.

Proposed System
In our project, it undergoes into two divisions: the first part focuses on grooming the face detection system based on the recognized human faces and detects face masks in the given bounded box area. The approach is by using Faster RCNN which is one of the architectures that uses Convolution Neural Networks ( CNN ). The other part of this project is to use a digital live recording camera at the entrance of any buildings to capture the photos of people visiting to work places or industries to improve their quality in-order to detect an individual face who is not wearing masks and acts as an input parameter to the face mask detection system.

Proposed Architecture
Initially, we take two types of dataset into consideration in our project which is nothing but a Kaggle Dataset and Customised Dataset. Kaggle Dataset consists of 853 images belonging to the people who wore masks, wore masks incorrectly and without masks under 3 classes. Moreover, it's been very much easier for annotating the dataset. Customised Dataset is nothing but a dataset which is randomly picked user images based upon various textures. Secondly, after preparing the dataset, labelling the images plays a major role. Under Labelling, the dataset is annotated by using the LabelImg tool. LabelImg tool is a graphical image annotation tool and it labels the objects inside bounding boxes in images Figure1 and

Detection of people without a mask
In this identification, we capture the faces, analyse properly in-order to get accurate results, and match the processed image with the stored database to retrieve the details of an individual who is not wearing the mask. Later on, with the integration of Twilio, it automatically sends and receives text messages to the respective people which intimates to wear a mask immediately. Twilio is nothing but a cloud based communication platform which automates those mentioned facilities with the help of web service APIs.

.2. Transfer Learning
Transfer learning is a technique where a pre-trained file is reused for another model on another job. At first, train a network on a base dataset, then reuse extracted features or shift into another network to get trained on a required data. This process will be suitable to work if both features of base and target tasks, instead of particular to the core task Figure 3.

TensorFlow API
The object detector using Tensorflow API is used in creation of a neural network that helps in reducing the problems for item classification. It consists of pre-trained files named as model-zoo. It has a variety of pre-trained files tested on the various data. It can be useful for initializing our program when executing on the ideal data Figure4.