AI Is Being Trained To Identify Faces In The Dark Using Thermal Images

Dhir Acharya


Night vision optics can help robots see in the dark, but there haven’t been any methods to train them to identify targets using just thermal imagery.

If this robotic invention comes to life, we could have killer robots with the ability to see and identify faces in the dark. Last week, the corporate research department of the US Army, DEVCOM, published a pre-print paper that documented how they developed an image database to train AI to recognize faces through thermal images.

Night vision optics can help robots see in the dark, but there haven’t been any methods to train them to identify surveillance targets using just thermal imagery. But that could change with this database, which includes hundreds of thousands of pictures of people in regular light and their thermal images.

An AI will be trained to use a number of parameters to categorize images. The artificial intelligence does not care if the picture of a face is in natural light or thermal images, it only needs a lot of data to recognize better. However, with nearly 600,000 images of only 395 subjects, the database is still small compared with standard facial recognition databases.

The insufficient data amount means that the AI won’t identify faces effectively. Researchers at DEVCOM conclude their paper:

“Analysis of the results indicates two challenging scenarios. First, the performance of the thermal landmark detection and thermal-to-visible face verification models were severely degraded on off-pose images. Secondly, the thermal-to-visible face verification models encountered an additional challenge when a subject was wearing glasses in one image but not the other.”

Ultimately, if the technology is further developed, it could result in better combat control on the battlefield. But if the results do not improve, it could lead to the death of innocent people that are falsely recognized.

>>> Company Uses Smart Seat Cushions For Staff Monitoring

Next Story