The US Army revealed its plan to develop robots that can see and identify faces in the dark with a facial recognition system.
DEVCOM, the US Army’s corporate research department, published a report about the development of an image database for training AI to perform facial recognition with thermal images.
The project could be a big step forward as robots will be able to use night vision optics in order to see in the dark. Until now, there had been no way to train them to identify surveillance targets using only thermal imagery. Indeed, the database the company has created is using hundreds of thousands of images including regular light pictures of people and their corresponding thermal images.
In order for it to be successful, the AI will be trained to filter these images using various parameters. With lots of data, the AI will then be able to get better at recognizing thermal images.
The database used is one of the largest to include thermal images, although it’s relatively small compared to standard facial recognition databases. Hence, this would mean that AI won’t manage to identify faces very well.
The researchers stated two challenges: the first one being the degradation of off-pose images by the thermal landmark detection and thermal-to-visible face verification models. The second one being the difficulties of the thermal-to-visible face verification models to recognize a subject when wearing glasses.