The outline of people’s bodies can be detected from Wi-Fi signals

Machine learning can analyse how the signals from Wi-Fi transmitters are disrupted by human bodies to reveal what position people are sitting, standing or lying in

By Alex Wilkins

A person’s bodily outline can be worked out from the pattern of Wi-Fi signals moving across a room, which could help monitor older people at risk of hurting themselves if they fall or enable the detection of home intruders.

Indoor cameras and radar-based technologies can be used to detect a person’s body shape or what position they are in, but these can introduce a privacy risk, such as a camera feed being hacked, or involve specialised hardware and installation.

Now, Fernando De la Torre and his colleagues at Carnegie Mellon University in Pittsburgh, Pennsylvania, have used indoor Wi-Fi to produce images of the shape of people’s bodies.

“People might be a little freaked out now, in the sense that internet service providers might locate what people are doing at home but, no, we are still not there,” says De la Torre. “The only thing that this paper shows is that, in a very constrained setting… [with] three receivers of Wi-Fi signal, there is enough signal there for the fine-grained detection of human body parts.”

To detect the body parts and where they are, De la Torre and his team placed three Wi-Fi transmitters and three receivers in different positions across a small room. When a person walks through the room, the presence of their body changes the pattern of the signal, which is recorded by the receivers.

The researchers then used a machine-learning algorithm, which had been trained on the relationship between Wi-Fi signals and visual images of a person’s body shape, to decode these recorded Wi-Fi signals.

“It’s like throwing out memory foam fibres into the whole room, and when we retract those fibres into a thread, our algorithm disentangles the fibres in the thread so that we can recover what the fibre has recorded,” says team member Dong Huang.

While the algorithm currently only works for the specific room and set-up it has been trained on, De la Torre and his team hope that it can be generalised to cope with more complex set-ups and real-world scenarios, where interference from other Wi-Fi networks might be present.

If it can be developed in this way, it could be used for cheap indoor monitoring to determine whether people in retirement homes have had a fall, for example, because it only requires two common Wi-Fi routers with three antennas each to replicate the number of antennas used in the study, or even to interact with computers, such as in an augmented reality set-up, says De la Torre.

Reference: arxiv.org/abs/2301.00250