The Defence Science Technology Organisation (DSTO) is running facial recognition trials which will underpin biometric initiatives across the Department of Defence, Immigration and new smartcard driver's licences.
The agency has toyed with so-called "attractors" including signs or noise-emitting devices that draw the attention of passers-by so they inadvertently look directly into a camera.
Angles of image capture is one of the biggest shortcomings of facial recognition devices, which often must be obfuscated yet be capable of taking a straight photo.
DSTO senior biometric systems analyst Brett McLindin told a Biometrics Institue conference today the trials will focus on the interactions between humans and biometric devices, rather than the technology itself.
"We are looking for generic results on human interactions, not just lab results. This will be used by the whole gammut of government," McLindin said.
Light and noise attractors were used to coax subjects to look into "pinhole" cameras, including one test which used an infrared beam placed before a doorway to trigger an alarm.
Subjects typically turned in the direction of the noise and looked directly into a camera. Another placed a camera in front of an illuminated sign which drew the attention of passers-by.
Initial trials were conducted between February and March this year with 316 DSTO staff and contractors. Some 1200 photos were taken of each participant including 13 different image types, using 14 cameras and five lighting scenarios.
Six trial types were tested including biometrics at a distance, identifying a face-in-a-crowd, and low light and night, and indoor and outdoor scenarios. In one trial, a specialised oscillating telescope was used for sub-pixel shift which produces better resolution.
McLindin told Computerworld it will take years for the results of the trial, which also includes PhD theses, to be finished and used by government departments, but noted sufficient raw data has been collected and tabled in 22 databases.
DSTO land and operations division and University of Adelaide psychology PhD student Dragana Calic is conducting trials on the human capacity for facial recognition to mimick real-world scenarios.
Under the four stage trial, the 316 participants were asked to choose a "look-a-like" photo from 1834 images of people sourced in the public domain including the Internet, while a third image was randomly chosen based on the subject's sex and ethnicity.
In upcoming stage in September, untrained human operators from the university will be asked to identify participants in a similar scenario faced by staff in charge of identity verification, alongside cognitive and personality assessments and followed by automated biometric tests.
The DSTO has called for researchers and vendors to provide specifications in biometric areas in need of development.