Rewrite this content
Installing cameras in a private place can cause privacy issues. A team of researchers from Carnegie Mellon University has found a way around this problem: they have developed a deep neural network that maps the phase and amplitude of WiFi signals received and sent by routers, which makes it possible to detect a human presence. and the movements of this one even though it is behind a wall.
Jiaqi Geng, Dong Huang and Fernando de la Torre are the researchers of this study published on arXiv titled DensePose from WiFi. They used DensePose, a system for mapping all the pixels on the surface of a human body in a photo, developed by Rıza Alp Güler from Imperial College London, Natalia Neverova, Facebook AI Research and Iasonas Kokkinos, University College London.
In 2015, researchers at MIT’s AI lab also used WiFi to detect people in another room: RF-Capture, their device transmits signals reconstructed in human form by an algorithm. Other researchers at the same facility had previously used the phone’s built-in wifi to see the number of people behind the walls, their gestures, and their precise location.
DensePose from WiFi
Advances in computer vision and ML techniques have enabled significant development in 2D and 3D human position estimation from RGB (Red, Green, Blue) cameras, lidars (laser remote sensing ) and speed cameras. But for researchers, estimating human pose from RGB camera images is affected by occlusion and lighting, lidars and radars are very expensive, beyond the reach of an average household or d ‘a little company.
Using WiFi is a good alternative for them to estimate human pose, cheap routers and humans interfere with WiFi waves. The waves, unlike the cameras, have no blind spots, and thanks to these interferences, they make it possible to see a person hidden behind a piece of furniture. Their model can estimate the dense pose of multiple subjects, with performance comparable to image-based approaches, using WiFi signals as the only input.
So they used DensePose and developed a deep neural network that maps the phase and amplitude of WiFi signals to UV coordinates in 24 human regions. The model differentiates humans from furniture in the room that it also detects.
WiFi human detection and privacy
For the researchers, their model could allow older people to continue to live in their homes in safety, protect privacy or “identify suspicious behavior at home”.
We can also imagine the opposite situation, malicious people to check that a room is empty or surveillance of our private life…
Sources : « DensePose From WiFi », arXiv:2301.00250
Auteurs : Jiaqi Geng, Dong Huang, Fernando De la Torre
Carnegie Mellon University
and add more content about WiFi and deep learning to detect people through walls