Army researchers attempt to match an infrared face biometrics against visible-light faces
Researchers with the U.S. Department of Defense (DoD) are experimenting with artificial intelligence and facial recognition to positively identify an individual hundreds of yards way in complete darkness.
The U.S. Army, always looking for a better way to identify specific people in the dark, says it is two years away from evaluating the technology, which uses infrared photography, in “operationally relevant environments.”
Researchers have already been working on the technology, called thermal-to-visible face synthesis, for five years. An algorithm compares a thermal image to a library of biometric photographs.
The system is about 90 percent accurate right now and could improve over time. It is not known if the system works with still images only or a video feed.
It is not known exactly where this work was begun by the Defense Department, but researchers with the Army Research Laboratory, University of Maryland and Brazil’s Federal University of Minas Gerais published a paper about thermal-to-visible face technology as far back as 2015.
It is possible that the Army’s approach could be overtaken by another method that came to light two years ago.
A research team at West Virginia University has proposed a method of identification people in the dark that involves capturing textural and geometric details found in visible light but absent in infrared images. The researchers used a coupled deep neural-network architecture that leverages large visible and thermal data sets. The network is trained by a polarimetric thermal face data set that they say is unique in the world.
In the private sector, OmniVision Technologies last June announced the launch of its new 2.9-micron pixel image sensor with 4-megapixel resolution to its Nyxel near-infrared and ultra-low light product line to support accurate biometric facial recognition at long distances.
The DoD is also looking into technology that can perform biometric identification through walls.