Related Experiment Video
Updated: Jun 27, 2026
06:37
Artificial Intelligence-Based System for Detecting Attention Levels in Students
Published on: December 15, 2023
Deep-Learning-Based Baseline Evaluation of Public WiFi CSI Datasets for Contactless RF-Based Human Activity
Tayyaba Parveen1, Rehan Khan1, Umer Saeed2
1Department of Electrical Electronic and Computer Engineering, University of Ulsan, Ulsan 44610, Republic of Korea.
Sensors (Basel, Switzerland)
|June 26, 2026
Summary
This study benchmarks deep learning models for WiFi sensing, finding CNNs efficient for structured activities and GRUs for dynamic ones. Results highlight challenges in distinguishing low-motion states and offer a reproducible framework for human activity recognition research.
Area of Science:
- Computer Science
- Electrical Engineering
- Signal Processing
Background:
- WiFi Channel State Information (CSI) is a promising technology for contactless human activity recognition.
- Inconsistent datasets and methodologies hinder reproducible comparisons of deep learning models in CSI sensing.
Purpose of the Study:
- To establish a unified baseline evaluation of four deep learning architectures (MLP, CNN, GRU, CNN-GRU) for human activity recognition using CSI.
- To provide a standardized preprocessing and training pipeline for reproducible CSI-based sensing research.
Main Methods:
- Harmonized multiple public CSI datasets and implemented a standardized pipeline for preprocessing and training.
- Evaluated MLP, CNN, GRU, and CNN-GRU models on diverse sensing tasks: single-person activity recognition, fall-risk estimation, multi-person occupancy classification, and localization-aware activity recognition.
Main Results:
- CNNs offer a good accuracy-complexity trade-off in structured activities; GRUs excel in temporal dynamics but incur higher costs.
- MLPs generally underperform due to limitations in capturing spatial-temporal dependencies.
- Dynamic and low-motion states remain difficult to differentiate, emphasizing the need for robust temporal modeling.
Conclusions:
- The study provides a reproducible framework and benchmark results for CSI-based human activity recognition.
- Future research directions include cross-dataset generalization, advanced hybrid model designs, and lightweight deployment strategies for WiFi sensing.
