Only utilizing WiFi signals, researchers managed to estimate the dense pose of multiple subjects with the help of an AI model. The output was comparable to a conventional image-based approach. The study seems to have happened in a lab environment with the subjects in between multiple specifically placed antennas. The authors admit that the model still struggles with limbs, with three or more people in one capture and with room layouts it has not seen during training. The basic idea is not new: MIT’s RF-Pose estimated 2D poses from radio signals in the WiFi frequencies back in 2018, even through walls, by training its network against a vision model.
Combining this research with the ToF localization approach as proposed by Wi-Peep might even provide more insights.