Facial Recognition that Sees Through a Mask
eTN Podcasts: www.livestream.travel
A groundbreaking artificial intelligence-powered facial recognition system that can identify people wearing masks has been deployed by law enforcement and intelligence agencies around the globe.
Developed by Corsight AI, a subsidiary of Tel Aviv-based computer vision company Cortica, the technology can also recognize individuals in extreme low-light conditions.
Using a reference image or video as a starting point, the system can ID people with as little as 40% of their face being visible, making it highly pertinent to the coronavirus pandemic.
“I see most of the players in the facial recognition market struggling with the COVID-19 masks, but our system was built from Day One to be able to recognize people from only part of the face,” Ofer Ronen, vice-president of business development at Corsight AI, told The Media Line.
“We were built to find a single terrorist within a crowd when he’s trying to disguise himself,” Ronen said. “So we don’t need a full face.”
The majority of facial recognition technologies currently on the market are not advanced enough to discern people’s identity when their face is partly covered. In March, Chinese company Hanwang Technology Ltd. announced that it had also developed a solution that can “see-through” the masks that many are wearing to stop the spread of the coronavirus.
Corsight’s system processes information taken from surveillance cameras, pictures and other visual sources to create a profile of an individual. Several of its researchers are former members of Israel’s 8200, an elite signal intelligence unit of the IDF.
Although the company completed development of its system just a few weeks ago, Corsight says it is already working with airports and various government bodies. In Israel, the firm is in the process of conducting a pilot test at an undisclosed hospital.
“Most of our customers we cannot disclose because they are intelligence agencies and special law enforcement units in different countries,” Ronen noted. “I can mention that we are deployed in several police units in Asia, Europe and even in Israel.”
Ofer Ronen (Courtesy)
When combined with a thermal-imaging camera, the system can help with COVID-19 contact tracing by identifying those with high body temperatures and flagging them for manual inspection.
Once a person is confirmed to have a fever, he or she is automatically added to a database that compiles all the locations the person visited that have surveillance camera footage. Those who came into close contact could then be alerted.
“If the body temperature is above 38° Celsius (100.4 °F), they are automatically [placed] in our system,” Gad Hayut, director of technical services at Corsight AI, explained to The Media Line.
“We combine this with the facial recognition technology,” he explained, “and then, whenever the camera sees [the individual], we will know that he was a threat at some point.”
What kind of data is stored? That is entirely dependent on the client and local regulations, according to Ronen, who also emphasizes that Corsight AI does not deal with the data side of the facial recognition equation. Rather, the client, such as a law enforcement agency, decides which kind of data to store, and where.
“We provide tools to allow high functionality while keeping the saved data to a minimum in order to support the need for privacy,” he specified. “There’s always a risk with such technology.”
Indeed, as such systems grow ever more powerful, some are concerned that facial recognition could be used for nefarious purposes by authoritarian governments to suppress entire populations. China, for instance, has already used this type of technology to racially profile Uighurs, a predominantly Muslim minority, according to Western media reports.
In order to address these concerns, Corsight AI has put in place a privacy advisory board made up of leading security and data-privacy experts. The panel is responsible for approving each business deal on a case-by-case basis.
“We will not sell to governments [if] we are not confident that they will not abuse the technology,” Ronen emphasized, adding that the goal is to “save lives” with the AI-powered system.
“It could be saving lives by finding a single terrorist at the airport, like the bomb attack in Belgium,” he said, referring to the 2016 Brussels bombings in which 32 civilians were killed and hundreds more injured in a series of coordinated attacks at the airport and a subway station.
“Or it could be used to save lives by recognizing a COVID-19 sick person in the crowd, [seeing] who he was talking to and checking these people,” he said.