Indoor observation becomes valuable. It can help provide for vulnerable individuals, such as the growing number of elderly people living alone. These systems can detect when people are falling or not moving at all, can detect intruders and provide peace of mind.
But this type of monitoring doesn’t always work and can be intrusive. Images from video cameras can be difficult to interpret when human bodies are partially obscured or when the light is poor. And few people want surveillance cameras in places like bathrooms and bedrooms.
A better option would be a non-visual way to monitor internal settings. But while radar and lidar systems can be effective, they tend to be too expensive for most budgets.
A non-visual image
Now, a significantly cheaper option seems possible thanks to the work of Jiaqi Geng and colleagues at Carnegie Mellon University in Pittsburgh. This team has found a way to monitor the movement of people indoors using WiFi signals. They even trained a neural network to use these signals to recognize the specific posture of each person in the room. “This paves the way for low-cost, widely available, and privacy-preserving algorithms for human sensing,” they say.
WiFi consists of low-power 2.4 GHz radio signals that encode data using various protocols. Geng and co ignore the encoded data and simply consider the ratio of the transmitted signal wave and the received wave. This depends on factors such as the distance between the transmitting and receiving antennas and any objects in between that may absorb, reflect or distort the signal.
A single transmitter and receiver obviously gives limited information about 3D environments. So Geng and co instead looked at signals from three transmitters and three receivers, pointing out that many commercial WiFi transmitters have three antennas and are thus well-tuned for this.
After cleaning the signals, Geng and co combined them to create a two-dimensional feature map that is similar to an image. But even though this image captures the pose of each person in the room, it’s not one that the human eye could easily understand.
So the researchers trained a neural network to recognize human poses instead. This is possible because a neural network called DensePose already does this in the visual domain. Geng and co applied this learning to the WiFi domain using a technique called transfer learning.
WiFi signals can reveal people’s posture in indoor environments (Source: arxiv.org/abs/2301.00250)
The result is a system that can recognize the posture of people in a room based solely on the WiFi signals passing through the environment. It’s not perfect though. The team says their system performs well at estimating the poses of human torsos, but struggles at detecting details like limbs.
It also needs to be completely retrained every time the environment changes – when it moves to another room, for example. “WiFi signals in different environments show significantly different propagation patterns,” say Geng and co. “Therefore, it is still a very challenging problem to implement our model on untrained layout data.”
This is interesting work showing the potential of using WiFi signals to monitor indoor environments, to detect falls or other potential problems.
However, it raises its own questions. WiFi signals are detected far beyond the boundaries of most homes and offices, and this increases the likelihood that malicious users will use this idea to monitor people’s activities behind closed doors. It instantly removes the privacy of walls and curtains, and although the resulting images are not ‘photographic’, this raises important questions of both security and privacy.
A recent criminal trend is the theft of high value vehicles by stealing radio identifiers from key fobs stored indoors. It turns out that for many cars, these signals can be remotely interrogated by anyone with the right radio equipment, who can then steal the vehicle.
Of course, there is a way to stop this kind of theft, and this has led to the appearance of tiny Faraday cells in homes. These are boxes lined with fine wire mesh that prevent radio signals from passing through and radio ID theft.
If WiFi imaging becomes commonplace, it certainly won’t be long before people start protecting their homes – especially their bedrooms and bathrooms – with similar but much larger cells.
That would be a shame, but perhaps a necessity.
Ref: DensePose From WiFi: arxiv.org/abs/2301.00250