Facial recognition was on the pinnacle of being accepted in certain applications as part of everyday life. But then the COVID-19 pandemic struck, and wearing a mask became the norm, but it also posed new challenges for facial recognition software. Facial recognition has been put to the test in two very different ways: monitoring the compulsory wearing of face masks in certain places as well as looking beneath the mask to identify faces.
Wearing a mask or face covering is not novel to most of the population of China. Since the SARS-CoV-1 outbreak in 2003, it´s been common to see members of the public wearing masks, although this has often been linked to avoiding the affects of pollution as much as stopping the spread of disease. In January 2020, the wearing of masks was made compulsory in Wuhan, followed by Shanghai in February, and other parts of the world as COVID-19 infections spread.
Other parts of the world (the West in particular) took longer to accept the new attire. In the US, around 35 states had imposed laws to mandate the wearing of a face covering by August 2020. However, according to a recent study, it´s believed that by the Autumn, up to 10% of the population were still ignoring the requirement.
US-based software solution provider, LeewayHertz, was quick to act. In April 2020, it launched its Face Mask Detection System, an app that uses CCTV, computer vision and AI to detect people who aren´t wearing a face covering. The app was designed primarily for use in airports, hospitals and offices, and alerts CCTV operators to staff and members of the public who are not complying with mask rules.
Furthermore, those behind developing software to recognize masks are confident that privacy regulations are still respected as algorithms don´t identify the person themselves, just the fact that they don´t have their nose and mouth covered.
Seeing beyond the mask
In another innovation prompted by the use of face masks, software is being developed, also through necessity, to see faces as if they weren´t wearing a mask at all. Anyone who thought they could get away with criminal activity by wearing a face mask will have to think again. And those traveling through airport and security checkpoints should soon be able to do so without having to remove their face covering. While in the early part of last year, many facial recognition systems were baffled by masks, developers have been working apace to train computer vision to see beyond the mask.
Traditionally, facial recognition has worked by measuring the distances between distinguishable facial features. We explained more about this in a news post of November 2019, ironically penned just as the COVID-19 virus was readying itself to change our lives completely. Covering your nose and mouth with a mask makes this pretty much impossible, and error rates in recognition of mask-wearers of up to 50% were reported in a NIST study in the middle of last year.
The good news is that software developers are working hard to adapt facial recognition to rely on the eye and forehead areas of a subject. In fact, some developers were working on this even before the pandemic, as the eye and eyebrow region are known to change the least over time, regardless of age or weight. And, as we´ve mentioned already, mask wearing has been common in some countries for quite some time. So it´s not surprising that one of the companies leading the way with the newest technology is Japan´s NEC (Nippon Electric Company).
NEC has been players in the facial recognition market for some time, and recently launched software that it claims is 99.9% accurate even when identifying a person wearing a mask. It´s worth noting, however, that a photo of the person without a mask is required for comparison, so this system will work well for border control applications but may be less useful in general surveillance.
UK-based TrueFace have also been working quickly to improve their facial recognition systems. Their technology is currently in use in two US air force bases to allow contact-free identification, a process that CEO Shaun Moore believes will be adopted more widely in public spheres such as sports and music venues. Their solution made it on to the top 10 list of tools published by the National Institute of Standards and Technology (NIST) and engineers are currently working to improve their error rate further in the case of mask-wearers.
Comparing Apples and pairs
And for those of you whose biggest inconvenience throughout all of this has been having to remove to your mask to unlock your iPhone, fear not, Apple is beta testing an iOS update to fix that. Phew! The downside is that you´ll need an Apple Watch as well to pair with.
Written by Natalie Ryan, Marketing Specialist at Active Silicon