Flaws in Face-Recognition at Palm Beach Airport
Data on Face-Recognition Test at Palm Beach Airport Further Demonstrates Systems' Fatal Flaws, ACLU Says
Tuesday, May 14, 2002
MIAMI--Interim results of a test of face-recognition surveillance technology obtained by the American Civil Liberties Union from Palm Beach International Airport confirm previous results showing that the technology is ineffective, the ACLU said today.
"Once again, a test of facial recognition by security professionals has shown that the technology is just not an effective way to increase our safety," said Randall Marshall, Legal Director of the ACLU of Florida.
"First there were lab tests, then experiments at the Super Bowl and on the streets of Tampa, and now at the Palm Beach airport. In every case, the experience has been the same: facial recognition is a clunker that holds little promise to make us safer.
According to documents released to the ACLU pursuant to a request under Florida's open-records law (the "Sunshine" law), the system failed to match volunteer employees who had been entered into the database fully 503 out of 958 times, or 53 percent of the time.
"Even with recent, high quality photographs and subjects who were not trying to fool the system, the face-recognition technology was less accurate than a coin toss," said Barry Steinhardt, director of the National ACLU's Technology and Liberty Program. "Under real world conditions, Osama Bin Laden himself could easily evade a face recognition system."
Seemingly disappointed airport officials also noted other limitations of the system. In order to work well:
- The subject could not be wearing glasses: "Eyeglasses were problematic," according to a summary of the test findings. "Glare from ambient light and tinted lenses diminished the system's effectiveness."
- The angle of the facial image could not vary: "There was a substantial loss in matching if test subject had a pose 15 to 30 degrees (up/down, right/left) off of input camera focal point."
- The subject had to be perfectly still: "Motion of test subject head has a significant effect on the system ability to both capture and alarm on test subject."
- The subject had to be properly lit: "System required approximately 250 lux of directional lighting" to work.
- The airport had to have high quality photographs. "Input photographs populating the database need to be of a good quality."
"It hardly takes a genius of disguise to trick this system," said Marshall. "All a terrorist would have to do, it seems, is put on eyeglasses or turn his head a little to the side."
The findings were based on the first four weeks of an eight-week trial. The test was conducted on about 5,000 passengers and employees a day at Palm Beach's Concourse C security checkpoint using a test group of 15 airport employees and a database of 250 photographs that, according to press accounts, features suspected criminals. There was no indication that the system successfully flagged any of those 250 suspects.
In addition, Steinhardt noted that the system's two to three false alarms per hour could prove debilitating if officials tried to expand the system to all passengers, since the number of errors would presumably expand as more photographs were added to the database.
"Once again, even in a pristine test using photographs of cooperative subjects taken under ideal conditions, face-recognition is a disaster," said Steinhardt. "We hope that Palm Beach County will recognize that this system is a waste of money and preserve scarce security resources for programs that will actually make us safer."
The Palm Beach report is online at http://archive.aclu.org/issues/privacy/FaceRec_data.pdf.
The ACLU has created a special report on face recognition technology, including a focus on Tampa's system, online at /Privacy/Privacy.cfm?ID=13484&c=130.