Focus face biometrics work on solving real-world problems, NtechLab Co-founder urges
The biometric facial recognition and video analytics system developed by NtechLab gives Moscow’s vast network of public cameras impressive capabilities such as high accuracy for people with different skin color and people wearing masks, and continuous identification through silhouette tracking, according to NtechLab Head of R&D and Co-founder Artem Kukharenko. The potential for these technologies in commercial settings, however, is just as great as for safe and smart city initiatives, Kukharenko told Biometric Update in an interview.
NtechLab’s biometric silhouette recognition went live on Russia’s massive CCTV network earlier this year, giving law enforcement a way to track criminal suspects across feeds from different cameras. The company has also developed an algorithm for analyzing the number of people in a space and their distance from each other, to support social distancing for the preventing COVID-19 transmission. Kukharenko describes the new features as part of the company’s focus on addressing real-life challenges.
“We are interested in area-wide cases where we work with a lot of video screens and so-called difficult scenarios, like different facial occlusions, different lighting conditions, different angles of view where people don’t look at the camera, they just pass near it,” Kukharenko explains. “For us its an interesting problem because it’s a real-life problem. We always try to improve the accuracy of our algorithm, and right now we have a leading position in these difficult wild scenarios.”
The algorithms must not only be able to handle real-life scenarios, they must be fast enough to deal with large volumes of data from camera networks, and small enough for the required hardware to fit the customer’s budget.
“Now we have optimized our algorithms so we can process more than 100 video cameras on just one GPU,” Kukharenko says. “It’s very important for our clients who have hundreds of thousands of video streams which need to be processed, and the total cost of ownership of the whole solution is very important.”
Shopping malls are another target market for NtechLab, where features like demographic analysis deliver value through marketing insights. Large-scale retail operations may also need a way to make data-driven decisions around traffic flow to preserve public safety, in which case silhouette tracking can collect the desired information without storing any sensitive personal information, according to Kukharenko.
“Our goal is to extract all the information from huge amounts of video streams, and to make some predictions on the data which our algorithms are able to extract,” he says. “That’s why we added to our facial recognition algorithm the silhouette tracking algorithm, age, gender, emotion recognition, and we also do car recognition so that we can track all the information which is available in the video streams and process it in some smart way.”
The most recent major partner announcement from NtechLab is that it is providing facial recognition for more than 1,600 schools in Russia. The system, dubbed ‘Orwell,’ is expected to be implemented in more than 43,000 schools, and indicates both the range of application areas, and the potential for controversy.
The company deals with the differing regulations and data protection laws in various regional markets it operates in through a policy of working with local partners to ensure both that each country’s laws are respected, and that the technology is applied correctly, Kukharenko says. This makes the quality of the local partner very important, but NtechLab also uses its technology to build preserving privacy functionality.
“We pay a lot of attention in our products to ensuring the privacy for clients,” Kukharenko states. “For example our software can be installed and run on the servers of our clients, so it doesn’t leave any external connection, and it can even be run without the internet so that it’s 100 percent secure, and the client can be 100 percent sure that the information doesn’t go anywhere outside the secure (perimeter).”
With a number of projects outside its home base of Russia in Latin America and the Middle East, and a growing presence in Europe and the U.S., attitudes towards facial recognition cover a wide range in markets NtechLab is active in. Kukharenko suggests that the key to public acceptance of face biometrics is knowledge, and organizations have to be prepared to explain what is really happening.
“If you look at history, in the beginning people are always a little bit concerned about every new invention. It was the same with mobile phones or with social networks, with 5G; it’s all the same,” he argues. “The better people understand how the technology works and what could be done with the technology, the less misconceptions there will be, and the better for society as a whole it will be. We believe in technology for good, and we know a lot of good applications of this technology, and a lot of areas where this technology could do a lot of good things and even save lives.”
Positive social impact is also being felt in less dramatic, but still important processes such as online banking, an area in which NtechLab has several pilots with large banks both in Russia and elsewhere, and retail goods delivery. Retailers have been forced by the pandemic to improve delivery systems, with facial recognition one of the technologies that can make sure the delivery reaches the right customer. Equivalent rethinking of business processes is happening in practically all fields, Kukharenko observes.
As adjustments are made in systems from retail and banking to public health and safety, biometrics companies have an opportunity to demonstrate the power facial recognition can have for good.
“We thought and spent quite a lot of time with our team of engineers trying to innovate in ways that can help in this crisis, and we developed our solution for preventing the (transmission) of the virus,” Kukharenko says. “Pedestrian detection and proximity detection algorithms are part of this system, and we implemented this solution in several cities.”
A focus on the issues faced by system operators in the field means NtechLab has put work into refining its algorithms for situations like difficult lighting or angles, high volumes of streaming video for real-time analysis, and accuracy with different races. Kukharenko says the company discovered a problem with its facial recognition’s performance with some demographics several years ago. NtechLab developed its own demographic tests, and then collected new training data to refine its accuracy to the point where it is no longer an issue, according to Kukharenko.
The change is a result, he says, of thinking constantly about how and where the algorithms will be used.
“It’s very important to tackle real-life problems, and not just spend time solving some academic problem if we are talking about production systems,” Kukharenko urges the biometrics industry. “In some cases academic problems are interesting by themselves, but if we want to apply something in real life then we need to solve a real-life problem.”