Researchers develop AI system able to see through walls
Researchers at the Massachusetts Institute of Technology’s Computer Science and Artificial Intelligence Laboratory have developed an artificial intelligence (AI) system that uses wireless signals to ‘see’ through walls. Called RF-Pose, the system uses a deep neural network to spot human motion using radio frequency (RF) signals reflected from people’s bodies. To create the system, the researchers collected, via a wireless device and a camera, examples of people walking, talking, sitting, etc. The images were then used to extract ‘stick figures’ which were fed to the neural network with the corresponding RF signal. Trained with this data, RF-Pose is able to estimate human postures and movements without the need for a camera, and only based on the wireless reflections that bounce off people’s bodies. According to the researchers, this radio-based system is almost as accurate as the vision-based system used to train in, and it can estimate 2D poses through walls without being trained on such scenarios. As explained by The Inquirer, the system could be used in search and rescue operations after natural disasters, and for medical purposes.