Face mask Detection using Deep neural networks

July 22, 2020 admin

Off

USE CASE:

This use case demonstrates the context of human by sputtering (spraying) transmitted virus, it appears necessary to wear the mask on the face to protect people and limit the spread of the disease. We are currently facing the Corona virus pandemic of 2019-20. Corona virus disease 2019 (COVID-19) is an infectious disease with initial, flu-like symptoms. COVID-19 first appeared in China, and then spread to the rest of the world quite quickly. The Contagiousness of COVID-19 is known to be high comparing it to the flu. In this approach, we propose a face mask recognition model that captures in real time if a person is wearing a face mask or not. Such application can be particularly useful for security purposes in checking if the disease transmission is being kept in check, especially for children and the elderly.

AITA APPROACH:

Face Mask Detection is an AI and Computer Vision driven image analytics solution which caters to the Covid-19 related violations. Its artificial intelligence program detects violations like Face Mask Detection, Social Distance Detection/This system can be deployed on the Hospitals, Office Premises, Government Offices, Schools and Education Institutes, Construction sites, Manufacturing units, Airports etc. The camera with AI-based face mask detection and social distance monitoring can generate real-time alerts. Face mask detection feature uses visible stream from the camera combined with AI techniques to detect and generate an alert for people not wearing face masks. A user-friendly interface allows monitoring and review of alerts generated by the system. In the fight against the corona-virus, social distancing has proven to be a very effective measure to slow down the spread of the disease.

IMAGE ACQUISITION:

As for the latest common face masks, two are available Applications closely related and different, namely, facial mask detection task and masked face recognition task. The task of detecting face masks must identify whether a person wears a mask as required. Masked face recognition task has to identify a person with a mask with the same identity. Data set requirements are different for each task. The former needs only masked face image samples, but the latter requires a dataset that contains multiple face images with and without a mask of the same subject. Relatively, the Face Datasets Recognition function is tougher to construct. In order to handle face mask recognition tasks, this paper proposes two types of datasets, including Face without mask (FWOM), Face with mask (FWM). The introduction of FWOM and FWM is shown below.

1. FWOM:

A python crawler tool is used to crawl the front-face images of public figures and normal people alike from massive Internet resources. Then, we manually remove the unreasonable face images resulting from wrong correspondence. The process of filtering images takes a lot of manpower. Similarly, we crop the accurate face areas with the help of semi-automatic annotation tools, like LabelImg and LabelMe. 90,000 images of the same 525 subjects without masks. To the best of our knowledge, this is currently the largest real-world masked face dataset.

2. FWM:

In order to expand the volume and diversity of the masked face recognition dataset, we meanwhile have taken alternative means, which is to put on masks on the existing public large-scale face datasets. To improve data manipulation efficiency, we have developed a mask wearing software based on Dlib library to perform mask wearing automatically. This software is then used to wear masks on face images in the popular face recognition datasets. This way, we additionally constructed a simulated masked face dataset covering 500,000 face images of 10,000 subjects. In practice, the simulated masked face datasets can be used along with their original unmasked counterparts.

AITA NEURAL NETWORK ARCHITECTURE:

DETECTED IMAGE SAMPLE:

MODEL PERFORMANCE:

CONCLUSION:

A method is designed for detecting if a person is putting on a face mask from a video selfie. Different analysis scenarios have been experimented using diverse types of conventional mask and varied acquisition conditions. The performance of the designed method relies on the efficiency of the exploited face and face-feature detectors. In the present study, wearing glasses had no negative effect. The use of rigid masks seems preferable because they reduce possibilities of wrong positioning on the face. For this latter, the designed prototype can particularly be efficient. Hence, a promising face mask recognition model has been proposed. A proof of concept as well as a development base is provided towards reducing the spread of COVID-19 by allowing people to validate the face mask via their webcam. Moreover, this self-checking of the mask wearing could be exploited by monitoring-related applications as a conformity attribute of mask wearing. Future works may investigate the development of highly robust detectors by training a deep learning model with respect to specified face-feature categories or to correctly and incorrectly wearer mask categories.