Spoiler alert: If you want to hide from cameras, you’d be better off buying a T-shirt with a print.
During the recent quarantine, we have gradually grown to accept mask usage and become accustomed to the idea of an always-covered face. We have learned to wait for 20 minutes after applying a cream so that it is completely absorbed to reduce the risk of skin irritation, and no longer touch the outer part of our masks. In addition, we are also increasingly ordering stylish masks from domestic designers or even making them ourselves. However, one thing is still a source of violent indignation. No, not a careless neighbor neglecting protection measures, but the fact that our own phones have ceased to recognize us.
FaceID systems used in many modern smartphones were not ideal even before the pandemic: many of us had to take our heads off our pillows to show the phone our entire face. Nevertheless, this did not cause any severe inconveniences. It’s quite another thing, however, when you are standing near a supermarket POS attempting to pay using your phone without any success. Until recently, iPhone suggested entering a password after several failed face recognition attempts. In the latest firmware, developers added the option to immediately confirm a transaction with a digital code. Android smartphones have a rather convenient Smart Lock function allowing you to automatically release the lock if, for example, you are in a "safe" place (for the duration of quarantine, you can specify the nearest store as such a place). However, the face recognition problem still remains unsolved: gadgets have not yet learned how to identify an owner under a mask.
Do not think, however, that since your own phone does not recognize you under a mask, then you are completely invulnerable to all other monitoring systems. City cameras are much more precise and evolved dramatically during the quarantine, when it was necessary to track and identify self-isolation regime violators. Of course, Chinese manufacturers pioneered development in this area: SenseTime, for example, claimed that its recognition mechanism captures about 240 face points (mainly around the eyes, nose and mouth). This minimizes potential error, even if a person is wearing a hat, mask and dark glasses.
Russian companies such as CROC are also working on this problem. Tatiana Pavlova, a video analytics project manager, said that a biometric algorithm captures 68 reference points, which makes it possible to "identify people with up to 98% accuracy." She adds that 30% of the face surface is enough for the system to do the job. 3D scanning is used in the company’s office, with employee faces being automatically scanned as they pass through a turnstile. The technology was also used during the pandemic.
Moscow City Court conducted an online session using CROC’s system. Usually, a person is identified by a judge using their passport. During the online session, video analytics helped with this. A neural network identifies a person in a field of view and searches for a match across a base of reference images of the session participants. If someone leaves the field of view or another person sits down in his/her place, an indicator appears on the judge’s screenTatiana Pavlova
Analytics Project Manager
By September, Moscow plans to launch a face recognition system in the city’s subway. All passengers will be checked against the federal wanted list, with identification being carried out using, among other things, a person’s iris and tattoos. However, this is not really something new: according to the Moscow government, 105,000 cameras throughout the city have already been connected to the FindFace system. In one of his interviews, Alexander Kabakov, a co-founder of the NtechLab software vendor, noted that although a mask hides the face better than a mustache or glasses, it is highly probable that the algorithm will still be able to identify you.
The system itself, of course, does not threaten law-abiding persons; on the contrary, in an ideal world, it will reduce the number of crimes and increase the investigation success rate. However, many human rights defenders are concerned that governments will misuse the technology. For example, the Moscow court recently rejected a complaint filed by Alena Popova, who considers using biometric data to be an invasion of privacy. This opinion is shared not only in Russia, but also in the United Kingdom, where a non-profit organization called Big Brother Watch fights for the fair use of surveillance technologies, including video analytics systems.
Some scientists unwittingly find themselves on the same side with human right defenders. Researchers at the Northeastern University of Massachusetts recently revealed something that can deceive the algorithm much better than any mask: a T-shirt with a print that you might think is just a pretty pattern. In fact, the drawing is so catchy and chaotic that it prevents the system from even capturing the face (so, advice to wear bright clothes could pay off), meaning that you are simply not considered to be a human by the system. The developers claim that the print blocks recognition in 63% of situations. According to Tatiana Pavlova, "the probability of being caught while wearing an invisible T-shirt is almost halved." A makeup featuring random dots and stripes can make you less visible to cameras, although you then risk attracting people’s attention. Some time ago, Grigory Bakunov, Director for Technology Adoption at Yandex, was studying the impact of such unusual make-up, but later abandoned the project.
If a face is completely unavailable, the person can be identified with the help of other types of biometrics, such as voice, gestures, gait, hand geometry, venous pattern, fingerprints, and ear shapeTatiana Pavlova
Analytics Project Manager
And, while someone’s gait is still not commonly used for biometrics analysis, hands and veins are already being used for identification in sensitive facilities. It seems that, in the near future, even when wearing a mask, "getting lost in the crowd" will no longer be possible.