Researchers at La Sapienza University in Rome have created a system called WhoFi that can identify people based on the unique changes their bodies make to Wi-Fi signals, The Register reports.
The system uses Channel State Information (CSI) — data about the amplitude and phase of the radio signal, which can be obtained from standard Wi-Fi equipment. Even when several people are moving at the same time, the model can identify them individually due to the specific effects of their bodies on the signals.
Unlike previous attempts, WhoFi does not require training on a specific user within each new access point — the system uses a transformer model that can generalize signals and adapt to new conditions without retraining.
The test used the public NTU-Fi dataset collected on the University of Singapore campus, where volunteers walked past 8 access points. WhoFi achieved 95.5% accuracy in the re-identification task. Previously, another system, EyeFi, showed results in the 75% range.
According to the authors, WhoFi could be used for camera-less access systems, office presence monitoring, intrusion prevention, and in healthcare settings for contactless patient monitoring. However, there are concerns that the technology could pose privacy risks, particularly if used without users' consent.
Details are provided in the scientific paper WhoFi: Deep Person Re-Identification via Wi-Fi Channel Signal Encoding, published on arXiv.