The US military has developed a face recognition system in the dark

[Jollier]

Scientists from the US Army Science Center have developed an algorithm for recognizing faces from images from a thermal imager. The research preprint is published on arXiv.

The developers have trained a convolutional neural network to compare images of people's faces taken with a thermal imager with ordinary photographs. In the infrared image, the neural network highlights the contour of the face, as well as its key elements - the mouth, nose and eyes. Based on this data, it creates a synthesized image, and then finds in the database the most similar to it. Tests have shown that the program already works with an accuracy of 85%.

The US Army Science Center spokesperson said the technology will be used during nighttime military operations. It will help to identify especially dangerous terrorists or dignitaries in the dark. The program can be used in real time by connecting to a thermal imager and a computer, or after completion of the operation.

According to the Next Web, the new recognition algorithm can be used in conjunction with another development of the US military - a thermal imager that sees through walls.