You can now detect people simply by using Wi-Fi signals.

When we talk about body detection and tracking, there have been many advances in recent years using radar, RGB cameras, or LiDAR sensors. However, researchers from Carnegie Mellon University have shown that it is possible to do this with devices that are now in almost every home in the world. We are talking about wireless internet routers. Yes, believe it or not It is now possible to get “full tight human body poses” using Wi-Fi signals as the only input method..

The work of Jiaqi Geng, Dong Huang, and Fernando De la Torre is very interesting as they analyzed the combination of WiFi signals with deep learning architectures to achieve their goal. So they developed a neural network that displays the phase and amplitude of WiFi signals in UV coordinates within 24 areas of the human body.

The results obtained were very astounding as they were able to detect several objects using the information coming from the antennas of the WiFi router. And their quality is not inferior to other solutions based on image capture.

Of course, we are not talking about technology that is already ready for commercial or domestic use. However, the progress made has laid the foundation for the development of body detection solutions that are not only much cheaper than existing ones, but also provide greater privacy.

Using WiFi Signals to Detect People

Researchers at Carnegie Mellon University sought to continue existing research on using Wi-Fi antennas to segment the human body and locate its hot spots. And the first results are notorious. But why use WiFi signals to detect and track people when other, better technologies have already been developed for this?

The authors acknowledge that developments in areas such as augmented reality and autonomous driving systems are very important. But they also explain that solutions like RGB cameras or LiDAR sensors don’t have much reach. because they are expensive. But not only that, but also they require the use of very specific equipment and consume large amounts of energy. On the other hand, wireless Internet routers are much cheaper today and already present in the daily lives of millions of people around the world.

One of the most positive aspects of using Wi-Fi signals for body detection is that It is not affected by the typical annoyances that affect camerasespecially indoors. This is the case with shadows, darkness, reflections, and even obstacles created by furniture or other objects.

According to experts, this new modality will be especially useful in scenarios in which it is necessary to detect people, but without compromising their privacy. One application could be health and care for the elderly.

[…] Most people find it inconvenient to have cameras recording them at home, and in some places (such as the bathroom) it will not be possible to install them. These issues are especially important in healthcare applications, which are increasingly moving from clinics to homes, where people are monitored with cameras and other sensors. It is important to address the aforementioned issues in order to better help an aging population that is the most susceptible (especially during COVID) and in which there is a growing demand to be able to live independently at home. We believe that WiFi signals can serve as a ubiquitous replacement for RGB images for human detection in certain cases.

Detail from a study by Jiaqi Geng, Dong Huang, and Fernando De la Torre.

A very curious method.

Using WiFi signals to detect and track a body required building a very interesting infrastructure. During their study, the study authors did not require tune overly complex equipment. Since they needed three transmitters and three wireless receivers, they implemented only two wireless routers with three antennas each.

Subsequently, they created a deep learning architecture capable of detecting human bodies and displaying them using Wi-Fi signals as the only input method. Not an easy process, given that the algorithm had to deal with scenarios that had multiple actors and possible obstacles.

The final results were amazing. The researchers shared qualitative comparisons of Wi-Fi-only body detection with synchronized images from cameras and made it clear that the technology has great potential.

Of course, the system is not perfect. Some typical errors they found were that the device was detecting poses that were not in the training data; which results in incorrect body parts being mapped. Whereas with the simultaneous presence of three or more people, the model based on Wi-Fi signals could not extract the greatest detail for each subject. However, the developers believe that these problems can be solved with additional training.

There is still much to explore in this sector, and the authors of the study are already planning the next phase: predict human body shapes in 3D using only signals coming from WiFi antennas.