The US military is developing a portable face-recognition device capable of identifying individuals from a kilometre away.
The Advanced Tactical Facial Recognition at a Distance Technology project is being carried out for US Special Operations Command (SOCOM). It commenced in 2016, and a working prototype was demonstrated in December 2019, paving the way for a production version. SOCOM says the research is ongoing, but declined to comment further.
Initially designed for hand-held use, the technology could also be used from drones. say it could be shared with law-enforcement agencies. This technology would enable people to be identified without knowing they were even on camera. Human rights advocates have expressed concern over its use.
Advertisement
Vibration problem
The device is being developed by Secure Planet, a firm in Arlington, Virginia, that produces long-range face-recognition devices based on digital SLR cameras with commercial face-recognition software running on a standard laptop.
These devices have a range of about 300 metres, and extending that distance isn鈥檛 as simple as adding a longer lens to the camera, because this increases noise from vibration. Atmospheric turbulence also becomes a problem because the air acts as an ever-changing distorting lens, an effect visible as heat shimmer on warm days.
The challenge for the new extreme-range recognition system is turning captured images into something clear enough for the software, which works best with passport-style photos.
How Secure Planet is achieving this is unclear, but an Australian Department of Defence team has an algorithm capable of unscrambling atmospheric turbulence to aid long-range face recognition.
Image clarity
Other researchers have experimented with using what鈥檚 known as a convolutional neural network to transform a series of blurred images into a single distinct one.
Neural networks can also help extract pictures from busy environments. In passport checks, the subject is stationary, looking at the camera and is well lit. In the field, faces may be at awkward angles, moving, badly lit and obscured by shadows, headscarves, sunglasses or other obstructions.
Again, machine learning systems can turn images taken in such situations into something closer to passport photos.
In principle, such techniques might work at a range of a kilometre, says Walter Scheirer at Notre Dame University in Indiana. 鈥淲e have not yet hit a serious fundamental limit to long-range facial recognition yet,鈥� he says.聽 鈥淲e haven鈥檛 pushed the optics or the algorithms to the limit.鈥�
However, Scheirer is sceptical that reliable identification is possible with current technology at such ranges. 鈥淚鈥檇 be interested to know what are the imaging circumstances,鈥� he says. 鈥淎re they doing this on a clear, sunny day with a co-operative subject?鈥�
Rasha Abdul-Rahim at human-rights group Amnesty International finds the technology troubling.
鈥淯nder the disguise that they will be more accurate, that they will be able to distinguish combatants from non-combatants, we are seeing a dehumanisation and a lowering of the threshold for the use of force,鈥� says Abdul-Rahim.
She says this may be a step towards fully autonomous weapons that find targets and attack them without meaningful human control. She calls for an informed public conversation, as well as transparency over the use of the technology and in particular how the face-recognition system works.
鈥淯ntil governments can demonstrate that it is in line with human rights laws, this type of technology should not be used at all, especially in a situation where an algorithm will be determining life and death,鈥� says Abdul-Rahim.
