BiometricsEvaluating Face Recognition Software’s Accuracy for Flight Boarding

Published 14 July 2021

Recent tests show that the most accurate face recognition algorithms have demonstrated the capability to confirm airline passenger identities while making very few errors. Facial recognition is currently part of the onboarding process for international flights, both to confirm a passenger’s identity for the airline’s flight roster and also to record the passenger’s official immigration exit from the United States.

The most accurate face recognition algorithms have demonstrated the capability to confirm airline passenger identities while making very few errors, according to recent tests of the software conducted at the National Institute of Standards and Technology (NIST). 

The findings, released today as Face Recognition Vendor Test (FRVT) Part 7: Identification for Paperless Travel and Immigration (NISTIR 8381), focus on face recognition (FR) algorithms’ performance under a particular set of simulated circumstances: matching images of travelers to previously obtained photos of those travelers stored in a database. This use of FR is currently part of the onboarding process for international flights, both to confirm a passenger’s identity for the airline’s flight roster and also to record the passenger’s official immigration exit from the United States.

The results indicate that several of the FR algorithms NIST tested could perform the task using a single scan of a passenger’s face with 99.5 percent accuracy or better — especially if the database contains several images of the passenger.

“We ran simulations to characterize a system that is doing two jobs: identifying passengers at the gate and recording their exit for immigration,” said Patrick Grother, a NIST computer scientist and one of the report’s authors. “We found that accuracy varies across algorithms, but that modern algorithms generally perform better. If airlines use the more accurate ones, passengers can board many flights with no errors.”

Previous FRVT studies have focused on evaluating how algorithms perform one of two different tasks that are among FR’s most common applications. The first task, confirming that a photo matches a different one of the same person, is known as “one-to-one” matching and is commonly used for verification work, such as unlocking a smartphone. The second, determining whether the person in the photo has a match in a large database, is known as “one-to-many” matching.