Many countries are increasingly using face recognition technology to ensure order and security. How good is this technology and how has it evolved?
Although authentication is based on many factors, according to some experts working at Chinese artificial intelligence companies, the latest monitoring systems are able to identify quite accurately. Their views are based on recent research on surveillance techniques outside the face recognition system. These include identifying a partially exposed face, gait and different physical characteristics. The face recognition is a technology that is being used but also attracts many conflicting ideas. In May 2019, the US city of Oakland banned the use of facial recognition software because of concerns that it would give false results.
In China, advances in face recognition are used by many cities to “name” and embarrass people, seek criminals among the crowd or verify the identity of passengers at the airport. . Using sophisticated algorithms, technology companies and other large Chinese enterprises application systems in retail, tourism and banking.
How does artificial intelligence work in the face recognition system?
Face recognition system can identify people from the image database, including stills and videos. Using deep learning – a branch of artificial intelligence – will increase the speed of face scanning because it learns more about the data it processes. These systems require huge amounts of information to be faster and more accurate.
These systems produce the so-called “unique face mark” for each object through reading and measuring from tens to thousands of “nodes”, including the distance between the face and the width of the nose. , eye socket depth. With the surveillance network, the identification system handles many characteristics, including height, age and clothing colors.
On the iPhone, its camera can analyze more than 30,000 infrared points and create 3D models of user faces. It is designed to adapt to changes in their appearance as they wear makeup, hats, towels, glasses, and contact lenses.
Wear a mask that makes technology identify “hands-on”?
According to Chinese experts, not really. The advanced face recognition system is capable of analyzing hidden parts of the mask, helping to narrow down the subjects and making accurate judgments.
The group of researchers at the University of Bradford (UK) agrees with this idea. They published a report that said facial recognition technology could determine 100% right even if the top, right half or 3/4 faces were exposed.
If the system is only trained to identify the entire face, its success rate will be reduced to 40% when showing eyes and nose, 60% when the lower half of the face is hidden. In contrast, the success rate of systems is trained to identify each part up to 90% if exposed eyes and nose, even with faces that do not show eyes and nose.
Some other modern technologies support face recognition.
Identify gait: China’s Startup Watrix introduces software that can identify a person from a distance of 50 meters, even if they hide their faces or turn away from the camera. Technology works by analyzing thousands of figures about a person’s gait to build a large database. It includes the body, the movement of the arm, the bow, or the word bowl. This technology does not require cooperation from the object.
Voice recognition: Technology is not new, is widely used as a smart assistant in cars and home appliances. The software is designed to recognize voice, remember personal characteristics and automatically respond. The system combines quite well with surveillance cameras to record sound and analyze user data better.
Laser heart rate: The US Department of Defense has developed a device called Jetson, capable of identifying a unique sign of a person from a distance of 200 meters or more with infrared lasers. Laser sensors are often used to automatically record a patient’s pulse, detecting changes based on blood flow. In contrast, Jetson uses a technique called laser vibration to detect surface movements caused by heart rate. The Pentagon confirmed Jetson could be 95% accurate in the right conditions and the future will improve. Its drawback is that it needs a large heart rate database.
The London Police’s face recognition system has a wrong rate of up to … 81%.
Don’t be surprised if you are suddenly caught while walking in the UK someday.
Face detection technology being tested by Metropolitan Police is said to have an error rate of 81%.
According to a study by Essex University, the system mistakenly identified four of the five innocent people as wanted suspects.
If sued to go to court, this system is likely to be included in the “illegal” category.
For the purpose of the synthesis, an independent report on the process of testing this service by London police, Peter Fussey and Daragh Murray was given by Essex University “unprecedented” access. 6 out of 10 trials conducted from 6/2018 to 2/2019.
The duo joined the officers working in LFR control rooms (Live Facial Recognition – face detection) and in the field; They also attend meetings and interviews, as well as planning meetings.
“This report is based on the process of intimate participation in related processes using face-to-face identification technology of the Metropolitan Police,” said co-author Fussey.
“The problems such as those related to the LFR process are subject to inspection, and the test results must be public, which is reasonable” – he added.
The main concerns of these researchers are very legitimate.
They affirmed that the Metropolitan Police did not receive a “clear legal authority” to use the LFR in accordance with domestic law, or did not care about factors such as the nature of infringement of the privacy of technology or use. Biometric processing technology.
In addition, the two researchers argued that the police were missing in pre-test planning and technological conceptualization, resulting in a range of consensus, legitimacy, and faith.
In the six trials evaluated, LFR technology produced 42 matches, but the study authors said only 8 of them were certainly correct.