Chinese Skynet: recognize the face of the person wearing the mask, 90% accuracy

Initially, face recognition technology is a quick and convenient way to unlock your phone. Over time, we develop this technology (and collect enough face data) for effective security purposes; many law enforcement agencies, border gates or airports have begun to apply new technologies.

The question is: How good is face detection technology to be so widely applied? Especially when each face is different, and one can wear everything on the face to avoid security cameras, such as sunglasses or masks.

The question is why Oakland and San Francisco banned the use of facial recognition technology to track crowded areas; partly because they fear the results may be inaccurate, leading to disastrous mistakes. But they will soon have to think again, when many Chinese artificial intelligence companies have just announced a new tracking system that has the ability to identify faces very well, even when the subject has a mask.

According to the study, to gain new success, computer scientists learn to validate the few exposed faces, besides analyzing both gait and recognizable body language. .

It is common for new technology to come from China: we still know the Skynet system's superficial face recognition capability is installed across China. It can detect a criminal in a crowd of hundreds, identify a person and know where they are moving as soon as they leave the airport , even capture the man who has been hiding Perennial.

Sitting back and talking to the press, surveillance system researchers in China share a little bit about the technologies behind Skynet's success.

What will artificial intelligence contribute?

Basically, the face recognition system will confirm a person's identity from a database of multiple images, including videos and photos. Using deep learning technology to speed up the face scanning process, a system will gradually learn a huge amount of data. The more data, the faster and more accurate the system will be.

These systems create something called "face", with each person having a different face – as well as the variety of fingerprints. It measures the distance between the eyes, the width of the nose, the depth of the eye sockets and the same details on the face. With a dense network of security cameras like China, in addition to face data, other factors such as height, age, clothing colors, etc. will also appear in the database.

Does mask cover your identity?

According to Chinese monitoring expert, not really.

Advanced identification systems will analyze (and predict) the details below the mask. This ability allows the monitoring system to reduce the number of people matching low, thereby easily identifying the identity of the person wearing the mask.

The new Chinese report has the backing of scientists from abroad, namely a team of working researchers at the University of Bradford. Last May, British scientists published a scientific report, announcing facial recognition technology that can achieve 100% accuracy with just one side, half face or 3/4 face.

The British researchers stated: on the system only trained to identify the entire face, accuracy will be reduced to 40% when only eyes and nose are seen, 60% when the lower half is obscured. Accuracy will skyrocket to 90% when the system is trained to add incomplete face recognition, achieving high results even when looking at the face without eyes and nose.

How well does the Chinese system detect faces hidden under a good mask?

Although camera networks and facial recognition systems are already present throughout China, their accuracy still depends heavily on having to see the whole face. Many Chinese people are willing to provide personal information, including their faces, to large companies, in exchange for using the service more easily. Companies also take advantage of data from customers to help improve technology accuracy.

As such, the majority of Chinese artificial intelligence companies get huge amounts of data from two main sources, users voluntarily provide and data from their own products.

A researcher pointed out that the successful recognition rate will decrease to 70% if the user covers only 1/5 of the face. However, the ratio will vary based on the partial cover.

At the present time, some Chinese AI companies are conducting system training with partially masked faces, in order to increase the success rate in all conditions to be reached.

What recent breakthroughs can help machines identify both covered faces?

The first is technology to recognize gait. New Startup Watrix introduced a software that allows you to identify a person's gait from a distance of 50 meters, even when they turn their back on the camera.

The system is trained by thousands of measurements related to a person's gait and body structure, from swinging his hands to walking. New technology allows people to try to hide their face away from the surveillance camera, by mask or by just turning away.

Second, it is voice recognition technology. It is getting better and better since the day people used voice as evidence to bring it to court; Now, people use voice to activate all kinds of systems. Using voice and camera surveillance is only a matter of late.

Finally, it is the laser system that reads the heart rate remotely. This is a technology developed by the US Department of Defense, with its own name, Jetson.

It can detect subject heartbeat from a distance of 200 meters, through an infrared laser system. It can work well when the subject wears ordinary thin clothes.

The Pentagon confirmed that Jetson is 95% accurate when used in the most optimal condition, and Jetson will become more and more effective in the future. However, for Jetson to identify a person, it is necessary to have a database of heartbeats first.

It can be seen that identification technology is more and more advanced. Starting with the face, we gradually got both gait and voice recognition, and then technology seems to be only in fictional movies: heartbeat recognition! The first three factors can be done by humans, but when we reach the limit named "heartbeat", we can see clearly how far and far away machines are.

Source : Genk