Criminal Faces in the Crowd Still Elude Hidden ID Cameras
When police used secret cameras to scan the faces of 100,000 people at the Super Bowl, they appeared to be fulfilling a longtime goal to rapidly identify criminals in a crowd.
But worries about the invasion of privacy may be premature because the technology is still far from foolproof and not yet widely used.
Experts warn that covert digital facial scans can be highly unreliable in public settings because digitized photos shot at an angle or in poor light create images that often fail to match existing mug shots.
This facial recognition technology is still a small industry, the systems are rarely used in public areas other than casinos, and even some law enforcement agencies are unclear about their effectiveness.
“The laboratory results suggest that this type of system has no chance in hell of working” for more than a rough filtering of suspects, said Jim Wayman, director of San Jose State University’s Biometric Test Center, a federally funded testing lab.
Indeed, a recent study found that digital comparisons of posed photos of the same person taken 18 months apart triggered false rejection by computers 43% of the time, according to the National Institute of Standards and Technology. A forthcoming Defense Department study substantially validates that finding, Wayman said.
“To these systems . . . one out of every 50 people looks like Carlos the Jackal,” an infamous terrorist, he said. “And the real Carlos has only a 50% chance of looking like himself.”
Yet companies selling facial recognition systems claim a high degree of accuracy and want to expand this visual technology to everything from monitoring automated teller machines to traffic checkpoints.
“In most big cities, you have hundreds of thousands of outstanding felony warrants and certain areas where a lot of people congregate. You can possibly pick them up that way if you can quickly identify them,” said David Watkins, managing director of Graphco Technologies Inc., which developed the surveillance software used at the Super Bowl. “You can identify known sexual predators that are lurking in a school area.”
At the Super Bowl in Tampa, Fla., police identified 19 people as having a criminal history by cross-checking their images against criminal files. News of the secret camera surveillance, first reported by the St. Petersburg Times, triggered a wave of complaints from privacy advocates concerned about Big Brother-type data banks.
In Las Vegas, some gambling houses already use facial recognition technology, but one software manufacturer warned that its effectiveness is compromised in casino environments.
“The lighting is weird, and the [camera] angles are acute,” said Iain Drummond, president of Imagis Technologies Inc. of Vancouver, Canada. “To recognize someone using facial recognition in a casino is a real challenge. When we installed a [trial] system at the [now-closed] Desert Inn, where the ceilings were 30 feet high, all we were seeing were the tops of heads,” he said. “There are relatively few examples of this being used to real success.”
To improve reliability, cameras must be made flexible enough to rotate 360 degrees to follow suspects, with the ability to zoom in quickly, Watkins said.
But casino operators other than Drummond claim better results. Three casinos owned by Donald Trump in Atlantic City have used facial recognition technology for 2 1/2 years and say it is a rousing success.
“In the first week after we installed it, we arrested a big group of cheaters,” said Charles Guenther, director of surveillance for Trump Marina Casino.
The system works even when gamblers look downward, away from the ceiling cameras, he said. “We just wait for an opportune time to get a good facial shot. . . . Once we get a good face shot, we download it. We just need one.”
Biometrica Systems Inc., headquartered in Mont Vernon, N.H., has sold its version of facial recognition systems to more than 100 casinos nationwide, said Jim Pepin, the company’s vice president of sales and marketing.
Pepin called the system “a kinder, gentler Big Brother.”
Tampa Tryout Is a Publicity Windfall
The nascent biometrics industry--which develops tools to measure physical characteristics such as vocal tones, faces, fingerprints and retinas to verify identity--is projected to bring in only $165 million this year worldwide, according to the International Biometrics Industry Assn. But backers hope that applications will be expanded to many commercial arenas, such as retail clothing store chains and other places where known shoplifters can be identified as they try to walk into a store.
“If you have certain people who are arrested in one store in one city, you could put them in a database, and when they walk into that store or another in a different state, you could tell when they walk in,” Watkins said.
Graphco offered its technology free at the Super Bowl to work out bugs before making it commercially available. The experiment certainly proved a publicity windfall for the small, 5-year-old company, which managed less than $6 million in sales last year.
The Graphco technology was originally developed for use in airports overseas by a country trying to track terrorists trying to enter, Watkins said. He would not identify the country.
Before the Super Bowl, Tampa police provided Graphco with a databank of 1,700 people who had been convicted of crimes ranging from ticket scalping and fraud to violent crimes, Watkins said.
As fans entered the stadium, hidden cameras scanned each of them. After matching the photos taken by surveillance cameras with those in the database, police identified 19 men, none with significant records, said spokesman Joe Durkin.
One man described as a longtime ticket scalper was picked out by the system, but when police arrived, “he had already fled,” Durkin said. No other suspects were apprehended.
This points up a problem with relying on facial-scan detection in crowded settings, said Doug Tygar, a professor of computer science at UC Berkeley. No system works fast enough to immediately apprehend a suspect before that person melts into the crowd. At an event like the Super Bowl, even a delay of a few seconds between when the photo is taken and the suspect is identified could make apprehension unlikely.
An FBI spokesman said the bureau has not used the facial recognition technology and could not comment on its effectiveness.
During the Democratic National Convention in Los Angeles last summer, federal and local authorities did not use any kind of cross-match program, said Frank O’Donnell, then head of the Secret Service’s Los Angeles field office.
In contrast, the use of video and closed-circuit TV cameras, O’Donnell said, is “everywhere, no question about it.”
Going Beyond Closed-Circuit TV
Many municipalities use such cameras to help police monitor crowds in commercial parts of town or to take pictures of people going through red lights--and then mail drivers a photo and a traffic ticket.
Cameras are prevalent at sports arenas such as Staples Center to allow security and police to watch for trouble. At the Arrowhead Pond of Anaheim, the home of the Mighty Ducks and a popular concert venue, closed-circuit cameras are in use 24 hours a day. General Manager Tim Ryan said virtually every new arena includes such cameras.
Indeed, at the Rose Bowl, “we did use some sophisticated counterintelligence for the [Women’s 1999] World Cup,” said Bruce Linsenmayer, Pasadena police commander for the unit that handles security. “We wouldn’t talk about it then, and we can’t talk about it now.”
As for the Super Bowl identity-scanning test, he said, “We would be interested . . . to see whether it’s applicable to us.”
Beyond specialized casino applications, few digital cross-matching products are for sale.
When those systems do hit the market, they will cost as little as $5,000, a small expense compared with multimillion-dollar video surveillance systems.
The ability to identify even a few dangerous criminals in public settings would make such an investment worthwhile, some police officials say.
“Any technology that can enhance public safety but not violate the individual rights of citizens is worth pursuing,” said Chief Michael D. Brasfield of the Fort Lauderdale, Fla., Police Department.
However, the covert nature of the process, which has sparked privacy concerns, also can be self-defeating, Tygar said.
“The advantage of [airport security] systems is that they make someone aware that their identity is being checked,” he said. “It helps address the privacy problem, and they act as a deterrent.”
*
Times staff writers Bill Shaikin, Steve Springer and David G. Savage contributed to this report.
(BEGIN TEXT OF INFOBOX / INFOGRAPHIC)
High-Tech Security
Law enforcement officials have long sought a technological fix to the problem of identifying suspects rapidly in a crowd. Facial scans that can be compared with computerized images of known criminals are one possible, but controversial, solution. Such a system was used Sunday at the Super Bowl in Tampa, Fla. Experts warn this system still has many bugs in it.
Scanning the Crowd
Cameras were used at each turnstile at entrances to Raymond James Stadium.
* Cameras continuously compared faces in the incoming crowd with a database of criminal records from local, state and federal agencies.
* 19 people were identified with criminal histories, none significant.
How Facial Scans Work
Facial-recognition technology translates spatial relationships between various parts of the face into complex mathematical patterns. Subtle features such as nose bridges, cheekbone structures and eye-socket shapes are used to develop a unique numerical template called a faceprint, which can then be stored and compared like a fingerprint.
One type of faceprint measures the size and shape of features around the eyes or the center of the face, such as the width of the nose or the distance from the nose to each eye. In adults, these features don’t change because of aging, weight or facial hair changes. Another type measures the difference between the various features of two overlaid images representing data from the entire face.
* The face has about 80 “landmarks,” but it takes only 14 to 20 to reconstruct a unique facial pattern.
* There are about 1.1 billion facial images in databases worldwide.
Researched by VICKI GALLAY/Los Angeles Times; photos by Graphco Technologies Inc. and Reuters
Sources: www-white.media.mit.edu/vismod/demos/facerec/system.html, Graphco Technologies Inc.
