Camera systems and facial matching software used to capture and verify identities of individuals in small groups traveling in small groups in a checkpoint-type setting demonstrated their effectiveness in testing last fall conducted by the Department of Homeland Security Science and Technology (S&T) Directorate.

The best combination of cameras and biometric matching algorithms identified more than 97 percent of the people traveling in small groups, which included two to four people, and took less than two seconds per person, S&T said yesterday. Where there were errors, it was usually due to the camera system and not the facial matching algorithm, S&T said.

“What we found was actually group size didn’t matter,” Arun Vemury, lead of the DHS S&T Biometric and Identity Technology Center, told Defense Daily earlier this month during a virtual interview. “The systems were consistently able to process people less than two seconds per person.”

The camera errors were either not detecting or taking a photo of everybody in the group, Vemury said. A number of the matching algorithms were able to match 99 percent of the captured images, he said.

In addition to testing against small groups, as opposed to just individuals as in past biometric rallies, the 2022 Biometric Technology Rally also created privacy requirements and required camera vendors to supply the best image captured of an individual to be processed by the matching algorithms, Vemury said in the interview, which was embargoed until S&T released the results of its 2022 Biometric Technology Rally.

The privacy requirement entailed having the camera systems only capture faces of individuals that opted to use a lane where the biometric capture systems were set up and ignore people in a nearby lane that chose to not to have their identities verified biometrically or bystanders, Vemury said. Test results showed less than 1 percent on average of the opt-out individuals or bystanders were photographed, S&T said.

“So that just shows that if you introduce these requirements, industry can respond,” Vemury said.

Vemury said the 2022 Rally was set up to be particularly challenging for the camera providers, both in terms of supplying the best image for the matching algorithms and meeting the privacy requirements. S&T originally selected six camera systems but given the challenges, one vendor dropped out before installations began and another during the set-up period, he said.

The 2022 Rally took place at S&Ts test facility in Maryland and included 575 volunteers representing 54 countries, including men and women, various demographic groups, and people with lighter and darker skin tones. During the rally, a camera system had to fit within a six-by-eight-foot area and capture photos of individuals as they walked through the checkpoint-type environment.

In addition to the four camera systems, S&T used 10 face matching algorithms, allowing for 40 facial recognition configurations. S&T isn’t releasing the names of the vendors whose technologies were used for the evaluation.

Of the 40 combinations, 17 met S&T’s threshold true identification rate, according to the test results. In instances where cameras obtained a high-quality image of a person, 31 combinations met the objective matching goal of greater than 99 percent. As for the privacy requirement, 29 combinations met the zero percent goal of capturing bystanders and persons that chose an opt-out lane.

Currently S&T doesn’t have plans for a 2023 Rally. The directorate is preparing for the Remote Identity Validation Technology Demonstration this year to challenge industry to demonstrate secure, accurate and fair remote identity validation technologies that can combat fraud when users apply for government services, open bank accounts, or verify social media accounts.