Wi-Fi signals can sense people: What the research actually shows

Wi-Fi does more than carry data: changes in radio signals can also reveal clues about people moving through a space. Research from Carnegie Mellon and Karlsruhe Institute of Technology examines different sensing methods, their capabilities and the privacy questions they raise.

Dji Banner Inside Article - 728x90

Sensing is different from a camera

Wireless signals interact with walls, objects and human bodies. Those interactions change properties such as the amplitude and phase of the signal reaching a receiver. Researchers use these changes to investigate presence detection, movement and body-pose estimation. The resulting representation is not the same kind of data as a conventional photograph recorded by a camera.

Carnegie Mellon researchers’ DensePose From WiFi work uses deep learning to map Wi-Fi data to dense correspondences on body regions. The model relates signal features to surface coordinates within 24 regions of the human body. This should not be described as every household router becoming a camera that sees flawlessly through walls without preparation: the experimental setup and training conditions determine the result.

Identity inference in the BFId study

Karlsruhe Institute of Technology’s BFId research takes a different approach from pose mapping. It focuses on matching movement patterns associated with the same person. Beamforming feedback information, or BFI, used from Wi-Fi 5 onwards, contains information about a connection’s radio environment. The study examines the privacy implications of using that information for identity inference.

The researchers created a dataset containing recordings from 197 participants. They reported 99.5% accuracy for BFI-based recognition in a particular training and test arrangement. That does not mean a system can automatically discover the name of any passer-by in every environment. Training data, the setting, recording duration and signal-collection conditions all matter when interpreting the performance.

Detection without carrying a phone

One point highlighted by KIT is that a person does not always have to carry a Wi-Fi device. Communications between other devices in the surroundings can be sufficient for human movement to affect the signals. The finding also illustrates the distinction between knowing a network password and drawing inferences from physical movement; these are not questions that can be reduced to the same security mechanism.

Camera-free presence and movement sensing is being investigated for applications including building automation and care. What data is collected, who can access it and how people are informed remain important questions. The studies point to the need to incorporate privacy alongside functionality when developing wireless sensing. Their findings should not be generalised into a claim that every router automatically performs the same surveillance functions today.

Sources and further information: DensePose From WiFi paper · BFId paper · KIT announcement.

Reporting: Erkan Veyseloğlu