Study: Wi-Fi can accurately identify human body

Researchers at the Karlsruhe Institute of Technology (KIT) have demonstrated that it is now possible to identify a person based solely on Wi-Fi signals in the environment. Unlike traditional methods, this approach does not require a person to have a smartphone or tablet connected to the network: routers or other Wi-Fi transmitters that work close enough to each other, interacting with each other. This is reported by Interesting Engineering.

How it works

According to the study, the new technology analyzes the interaction of radio waves between standard Wi-Fi devices and forms a very detailed image of a person’s presence, posture, and movements – something like a camera image, but made up entirely of radio signals.

What’s the risk?

Professor Torsten Strufe from the Institute for Information Security and Reliability (KASTEL) explained that radio waves can reproduce images of people and spaces in the same way that a camera works with light. He stressed that the system is able to detect a person even if they do not have their own device, using active Wi-Fi networks nearby. This means that a person’s presence and movements can be recorded without their knowledge.

Another researcher from KASTEL, Julian Todt, warned that any Wi-Fi router could potentially become a surveillance tool. For example, if a person regularly walks past a café with the network turned on, their movements could be recorded and used by government agencies or commercial companies.

Strufe added that while there are simpler methods of surveillance, such as CCTV cameras, the widespread use of Wi-Fi poses the risk of an almost comprehensive surveillance infrastructure.

Unlike previous approaches that used LIDAR sensors or channel state information (CSI), which describe in detail how a radio signal travels from a transmitter to a receiver, the new technique uses only standard Wi-Fi devices and does not require specialized equipment. It is based on the analysis of beamforming feedback signals (BFI), which are transmitted in an unencrypted form. This data allows you to form images of objects from different angles, which are then processed by machine learning algorithms.

According to experts, AI models trained in this way will be able to recognize a person in just a few seconds.

It is noted that in an experiment involving 197 people, the researchers managed to achieve almost 100% accuracy in identifying a person regardless of their gait style or direction of gaze. This indicates serious risks to privacy: using this technique, a person can be recognized quite quickly, imperceptibly, without expensive equipment.