By: Parth Rajput, RIG Inc Researcher
Over the last year and a half, Covid has changed the way people interact with one another. While interpersonal communication over digital platforms is on the rise, we can expect that activites will again return to normal, albeit with some new norms in the foreseeable future at a global level. This provides us with a positive outlook but it is also an indicator of various concerns. Governments globally have been implementing various policies to ensure that people can return to the “old” social norm smoothly. To facilitate this vision, various AI/ML-based solutions are being developed and in this blog, we will discuss a few which are currently being researched.
Before we discuss the solutions, let us first understand some of the challenges being faced globally. As mentioned on the CDC Covid-19 information page, Covid spreads through breathing in the air around infected people, coming in contact with droplets/particles from sneezes/cough containing the virus, and lastly through touching eyes/nose/mouth with infected hands. To mitigate the spread, people are advised to wear masks in public spaces, sanitize their hands, and maintain social distance. In addition to these precautions, vaccinations are also recommended. Of these preventive measures, maintaining social distancing, and using proper sanitization methods need to be implemented at the individual level, whereas verifying an individual’s vaccination details can be achieved using simplistic mobile applications.
This leaves us with enforcing the most important preventive measures – How do we ensure people wear masks in public spaces? As social interactions go back to the old norm, for example ensuring that masks are being worn may be necessary. The simplest way to enforce this is to have someone monitor the venue’s entrance and ensure that masks are being worn properly. This then brings us to the next challenge: How do we ensure that people will continue to wear masks once they have entered the venue? Do we ensure that authorized personnel monitor patronsto enforce this? This is being implemented at many venues, making use of existing surveillance systems to autonomously detect and raise appropriate alerts is much more efficient.
Most public venues have an entire network of security cameras. There are typically cameras at the entrance and many more within the premise. These systems can be used to detect people and their masks using AI and computer vision-based object detection systems. Such a system can be used to detect a person and their mask, through training. If the person lacks a mask, the venue’s authorized personnel can be notified and appropriate protocols can be initiated. Similarly, the surveillance systems within a venue can be used to monitor masks within their range of reconnaissance and similar alerts can be sent.
Although this is an abstract representation of what an ideal system should look like, let us take a brief look at some of the efforts made to achieve these or similar results. In 2020, an article published by Oumina et al., explored the use of transfer learning for mask detection and compared it to various other implementations such as Convolutional Neural Networks (CNNs) to evaluate their performance. Rahman et al. similarly explored standard object detection-based techniques for this task by exploring CNNs for this purpose. CNNs have generally been used for object and face detection because of their ability to extract features directly from images. As evident from the two implementations mentioned, CNNs with various configurations can be tested to create the next benchmark model that gives state-of-the-art performance on this problem statement. In their blog post, Nvidia has elaborated on their healthcare-specific application framework, Clara guardian , and its ability to detect masks within healthcare facilities. In the last few years, other deep learning models have been explored and developed as alternatives to CNNs, to achieve greater performance. Some of these techniques involve the use of 3D Modelling techniques and Deep Neural Networks. We can also extend the abilities of Generative Adversarial Networks(GANs) to frontalize/ change facial poses to an acceptable angle and then use this for mask detection.
Revolutionary Integration Group Inc. provides various solutions that implement a person’s trust level dynamically using the Dynamic TrustTMmatrix. This can potentially be extended to incorporate aspects like Covid vaccine information along with the person’s masked frontal image taken from a smartphone. This can ensure that the person enters the venue with a mask and the person’s trust level can then be dynamically adjusted along with the trust of the venue.
While the problem statement is relatively new, there is a huge potential to improve and explore solutions in various directions to ensure public safety during Covid. One area of improvement, which is discussed in reference 3 is the lack of an elaborate dataset. This also raises the possibility of testing out existing objects and face detection systems on a newly created image dataset comprised of images of people wearing masks. This challenge is not only an eminent need, but it also has the scope of being expanded to various other applications, such as ensuring proper lab gear is being worn in research laboratories, surgical rooms, and hazardous work environments. It is safe to assume that in the upcoming months, we can expect a lot more research and solutions in this domain.
 A. Oumina, N. El Makhfi and M. Hamdi, “Control The COVID-19 Pandemic: Face Mask Detection Using Transfer Learning,” 2020 IEEE 2nd International Conference on Electronics, Control, Optimization and Computer Science (ICECOCS), 2020, pp. 1-5, doi: 10.1109/ICECOCS50124.2020.9314511.
 M. M. Rahman, M. M. H. Manik, M. M. Islam, S. Mahmud and J. -H. Kim, “An Automated System to Limit COVID-19 Using Facial Mask Detection in Smart City Network,” 2020 IEEE International IOT, Electronics and Mechatronics Conference (IEMTRONICS), 2020, pp. 1-5, doi: 10.1109/IEMTRONICS51293.2020.9216386.