Szczegóły publikacji
Opis bibliograficzny
GaitSig: scalable human identification based on Wi-Fi CSI-extracted gait features / Michał MAJ, Katarzyna KOSEK-SZOTT // IEEE Access [Dokument elektroniczny]. — Czasopismo elektroniczne ; ISSN 2169-3536 . — 2026 — vol. 14, s. 127383–127396. — Wymagania systemowe: Adobe Reader. — Bibliogr. s. 127395–127396, Abstr. — Publikacja dostępna online od: 2026-08-11
Autorzy (2)
Słowa kluczowe
Dane bibliometryczne
| ID BaDAP | 170327 |
|---|---|
| Data dodania do BaDAP | 2026-10-01 |
| Tekst źródłowy | URL |
| DOI | 10.1109/ACCESS.2026.3722568 |
| Rok publikacji | 2026 |
| Typ publikacji | artykuł w czasopiśmie |
| Otwarty dostęp |  |
| Creative Commons | |
| Czasopismo/seria | IEEE Access |
Abstract
This article presents person identification system based on Wi-Fi channel state information (CSI) analysis. The proposed framework transforms raw CSI measurements into Doppler spectrograms and extracts gait-related features using a deep neural network. The network employs dilated convolutional blocks with attention mechanisms and BiLSTM layers and is trained using supervised contrastive loss to improve discriminative representation learning. Human classification is performed by comparing embeddings to per-person cluster centroids, which creates unique gait-derived signature for each human. Applying a Kalman filter at the network output significantly improves classification accuracy. It uses parameters obtained using the particle swarm optimization algorithm. The final improvement in gait classification is the use of a support vector machine, which allows for non-linear cluster separation. Unlike conventional closed-set approaches, the proposed framework enables the addition of new users without retraining the neural network, requiring only lightweight classifier parameters adaptation. The system was evaluated using two public datasets, Widar3 and EIGait, comprising 19 subjects recorded in six different environments, achieving an accuracy of 86.5% using the Kalman filter and 98.4% with a Kalman filter combined with SVM. The results indicate that KF-assisted neural networks are a promising direction for practical and scalable Wi-Fi sensing-based biometric systems.
