Real-Time, Non-Intrusive Fall Detection via Wi-Fi CSI: A Comparative Study of CNN-LSTM, GNN, and Transformer Models
AI, Computer Science and Robotics Technology · 04.12.2025 · 4
DOI: 10.5772/acrt.20250099 ↗Kurz gefasst
Fachartikel in der Zeitschrift AI, Computer Science and Robotics Technology, online erschienen am 04.12.2025. Verfasst von Dit Preechakarnjanadit und 1 weiteren Autor:innen. Die Arbeit nennt 25 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Falls pose serious risks, particularly for the elderly, and existing monitoring methods like CCTV and wearables face privacy, comfort, and setup challenges. This study introduces a non-intrusive fall detection system using Wi-Fi Channel State Information (CSI) for Human Activity Recognition (HAR). The dataset was collected from two participants across multiple sessions and environments, comprising 700 falling events and 2,700 normal activity events (lying down, staying still, transitioning, and walking). We evaluate Convolutional Neural Network–Long Short-Term Memory (CNN-LSTM), Graph Neural Networks (GNN), and Transformer models on CSI data, with and without Principal Component Analysis (PCA). CNN-LSTM achieves the highest accuracy (94.85% with PCA) and performs robustly in real-time. It is validated across five activities—falling, lying down, staying still, transitioning, and walking—under both Line-of-Sight and Non-Line-of-Sight conditions, including an extreme case with the router and device in separate rooms. Four-class accuracy reaches 95.5%, with PCA helping reduce confusion between similar postures. The system uses a sliding window, real-time tcpdump capture, and a multithreaded Python pipeline, achieving average latency of 2.3 seconds. To evaluate model reliability under limited resources, the models were trained on data from one participant and tested on the other two participants, providing a proof-of-concept for generalization to unseen subjects. This low-cost, privacy-preserving solution suits smart rehabilitation and elderly care.
Zitieren
APA
Preechakarnjanadit, D., & Taparugssanagorn, A. (2025). Real-Time, Non-Intrusive Fall Detection via Wi-Fi CSI: A Comparative Study of CNN-LSTM, GNN, and Transformer Models. AI, Computer Science and Robotics Technology, 4. https://doi.org/10.5772/acrt.20250099
Vancouver
Preechakarnjanadit D, Taparugssanagorn A. Real-Time, Non-Intrusive Fall Detection via Wi-Fi CSI: A Comparative Study of CNN-LSTM, GNN, and Transformer Models. AI, Computer Science and Robotics Technology. 2025;4. doi:10.5772/acrt.20250099
BibTeX
@article{Preechakarnjanadit2025,
author = {Preechakarnjanadit, Dit and Taparugssanagorn, Attaphongse},
title = {Real-Time, Non-Intrusive Fall Detection via Wi-Fi CSI: A Comparative Study of CNN-LSTM, GNN, and Transformer Models},
journal = {AI, Computer Science and Robotics Technology},
year = {2025},
volume = {4},
doi = {10.5772/acrt.20250099}
}
EUVIMED