MIT researchers have developed a vision system that allows to see a body through the walls using radio waves and a neural network to infer the person’s postures and movements. The envisaged applications are above all medical.
The Massachusetts Institute of Technology (MIT) has been working for several years on a system to see through the walls using the reverberation of radio waves. Called RF Capture, this amazing technology is not intended to serve as a spy tool, but rather to detect the falls of elderly people at home or to follow patients suffering from serious diseases (Parkinson’s disease, multiple sclerosis muscular dystrophy) or looking for victims under rubble after an earthquake.
The team in charge of this project has made further progress by associating its process with an artificial intelligence. Now, the new system, renamed RF-Pose, is able not only to distinguish the silhouette of a person through the walls, but also to deduce the person’s posture and movements.
The program materializes what it “sees” through the walls in the form of an animated skeleton that reproduces in real time the movements of one or more people. RF-Pose also works in dark environments. To achieve this result, researchers created a deep learning neural network that was trained from thousands of radio wave profiles and images corresponding to these profiles showing postures (sitting, walking, the opening of a door…). It was this combination that allowed the AI to learn to associate a radio signal with a movement.
Once trained, the algorithm worked only from radio waves. In addition to motion detection, RF-Pose has also proven to be very effective in accurately identifying someone in 83% of the cases on a list of 100 people. For the moment, the program generates a representation in 2D. But researchers are working on 3D modeling that would detect micro-movements, such as tremors in an elderly person.