Argus: Wi-Fi Telemetry Achieves 84.85% Accuracy in Identifying 154 Subjects

Argus, a novel passive Wi-Fi sensing system, can recognize individuals without the use of cameras, wearables, or specific movements. Researchers developed this technology, as outlined in a paper on arXiv (2608.14670). Argus utilizes standard Channel State Information (CSI) to generate compact statistical representations known as 'statgrams' from brief CSI intervals. A lightweight decoder-only Transformer interprets these statgrams as tokens, while segment-level logit aggregation compiles evidence over time. In experiments with 154 participants, Argus achieved a Top-1 accuracy of 78.88% ± 1.62% for 6-second intervals and 84.85% ± 1.31% after aggregating 19 overlapping windows within a 60-second timeframe. The evaluation employed a rigorous physical-segment split to ensure distinct training and testing data. This innovative approach presents a scalable, device-free alternative to conventional biometric techniques, with potential applications in security, smart environments, and personalized services. The paper's cross-type submission on arXiv suggests it may have been presented at a conference or journal, emphasizing the promise of Wi-Fi telemetry for person identification beyond just gait and activity indicators.