Agency aims to ID from afar

Agency aims to ID from afar

DARPA's systems unit teaches computers long-distance facial recognition

By Patricia Daukantas

GCN Staff

DARPA's Face Recognition Technology database of thousands of photos is the test bed for recognition systems such as this one developed at MIT's Media Laboratory.

Recognizing people from a distance is still something the human eye and brain do better than computers. The Defense Advanced Research Projects Agency wants to teach them how.

So-called multimodal computer technology can identify persons by recognizing faces, iris patterns, ear shapes and facial vascular patterns, said David Gunning, a program manager at DARPA's Information Systems Office in Arlington, Va.

Gunning described a just-approved multimodal study at a September meeting of the Biometric Consortium, a network of government agencies, vendors and academics interested in biometric identification and verification.

Human identification from a distance, known as HID, is important in guarding against terrorists who might be scoping out target buildings, Gunning said.

Although surveillance cameras are common, manual review of the video images is highly susceptible to error. An HID computer system could catalog repeat visitors automatically and detect known or suspected criminals.

Facial factors

HID, however, does not work well when subjects walk around casually and do not pose for an identification scan, Gunning said. The system must be able to distinguish the face from the background, to account for varied lighting and angles, and to discount eyeglasses, aging and other variables.

DARPA surveyed biometrics experts about measurable traits that are universal yet unique to each person and reasonably permanent over time.

Each identification mode has strengths and weaknesses, Gunning said. For example, iris scanning ranks high in claimed uniqueness and permanence, but at a distance, it runs into problems with focus, reflections and interference from eyelids and lashes.

Ultimately, an HID system probably will incorporate several technologies including face, iris and voice recognition, Gunning said. Less-developed technologies such as thermography and ear identification could supplement them.

Thermography measures the infrared radiation given off from a person's head. Humans have a 5-kilometer-long network of small facial blood vessels, and each individual's unique pattern of vessels and branch points exists at birth and grows predictably.

'I don't think anyone else has suggested ear identification,' Gunning said. The structure of the human ear does not change over time, and ear shapes can be detected in infrared or visible light.

Much work has already been done on voice recognition, Gunning said, but DARPA will emphasize recognizing voices in a natural environment with background noise and subjects who are not necessarily cooperative.

Gunning suggested a staged strategy, in which physique and gait measurements might first draw the HID system's attention to a distant subject. Face recognition and thermography could provide midrange evidence. Closer up, the detailed identification could be made by face and voice recognition, iris scans and even body odor.

Although Gunning presented a timeline of studies and experiments over the next five fiscal years, he called it 'notional' and invited suggestions from the biometrics community.

When DARPA is ready to solicit research proposals, the invitation will appear in Commerce Business Daily and on the agency's Web site, at

Stay Connected

Sign up for our newsletter.

I agree to this site's Privacy Policy.