NIST testing suggests best face biometrics algorithms for flight boarding already meet accuracy goal
Some vendors already meet the U.S. governments standards for facial recognition accuracy in relatively challenging biometric airplane-boarding scenarios across all demographics, according to the latest testing from the U.S. National Institute of Standards and Technology (NIST).
A new Face Recognition Vendor Test (FRVT), ‘Part 7: Identification for Paperless Travel and Immigration,’ considers face biometrics for a one-to-many scenario with a limited database, such as would be used in double-duty for flight boarding and departure facilitation; access control and record-keeping applications, respectively. It covers different possible points of failure in the biometric matching process, from database creation through presentation attack detection.
In testing against databases made up of 420 or 42,000 travelers, NIST found that using multiple images of the same person significantly decreases errors.
The vendors’ biometric algorithms were tested with a false positive threshold of 1 in 3,333 (0.0003 percent).
The most accurate algorithms have insignificant variations in accuracy between different demographics, and the top-performing algorithms searching against 42,000-image galleries were below 1 percent. Searching databases of only 420, error rates were often three times lower.
Out of 567 simulated flights with 420 passengers, as many as 428 can be completed without any false negatives, according to the test results, or 99.5 percent of passengers can go through the biometric boarding check successfully on the first attempt, using any one of seven algorithms. The most accurate, however, could successfully board 545 of 567 flights.
At least 18 different developers submitted algorithms effectively boarding 99.5 percent of travelers in the test scenario. Nearly all developers easily pass the 3 percent false negative goal “implied by legislation in the U.S.” as an average, even with single-image enrollment, but broken down by demographics, some algorithms do not appear up to the task.
For the larger-gallery tests, only some of the tested algorithms meet the standard: “(A)ll algorithms except some from Cloudwalk, Deepglint, Idemia, NEC, Paravision, Sensetime, Visionlabs and X-ForwardAI cannot maintain FNIR below 0.03 so, depending on FPIR, would not be meeting a legislative mandate for FNIR<0.03 and TPIR>0.97,” NIST writes.
Market still booming
The market for airport technology is expected to continue growing rapidly, with facial recognition among the drivers, according to a Global Industry Analysts report that puts the overall ‘Global Smart Airports’ market at $7 billion in 2020, reaching $27.4 billion by 2027, a 9.3 percent compound annual growth rate.
A growing number of airports across Southeast Asia are trialing artificial intelligence for computer vision to increase airport efficiency and deal with passenger volumes, Graymatics CEO and Founder Abhijit Shanbhag writes for Finance Digest.
Shanbhag refers to the deployment of facial recognition in Malaysia’s Kuala Lumpur International Airport, and face and iris biometrics at Singapore’s Changi Airport as examples. The trend is likely to continue, he writes, between the benefit of reduced touchpoints, and possibly in combination with “digital vaccine passports.”
It is deployments of SITA’s biometrics for touchless processing at Bahrain and Istanbul Airports that the company’s European president, Sergio Colella, refers to an in interview with Simple Flying. Similar to Shanbhag Colella sees continued proliferation of contactless systems and increasing integration across the airport environment.
accuracy | airports | algorithms | biometric identification | biometrics | contactless | facial recognition | matching | NIST | passenger processing | presentation attack detection | standards | testing