Related Experiment Video
Updated: Aug 30, 2025
05:57
Long-term Video Tracking of Cohoused Aquatic Animals: A Case Study of the Daily Locomotor Activity of the Norway Lobster Nephrops norvegicus
Published on: April 8, 2019
6.9K
HHI-AttentionNet: An Enhanced Human-Human Interaction Recognition Method Based on a Lightweight Deep Learning Model
Islam Md Shafiqul1, Mir Kanon Ara Jannat1, Jin-Woo Kim1
1Department of Electronic Engineering, Kwangwoon University, Seoul 01897, Korea.
Sensors (Basel, Switzerland)
|August 26, 2022
Summary
This study introduces HHI-AttentionNet, a lightweight deep learning model for recognizing human-human interactions using WiFi signals. The model achieves high accuracy with fewer parameters, improving upon existing methods for WiFi-based activity recognition.
Area of Science:
- Computer Science
- Artificial Intelligence
- Signal Processing
Background:
- WiFi-based human activity recognition (WiFi-HAR) offers privacy and cost-effectiveness for indoor sensing.
- Recognizing human-human interactions (HHIs) using channel state information (CSI) remains a challenge.
- Existing deep learning (DL) models for HHIs often have high parameter counts, limiting their use on resource-constrained devices.
Purpose of the Study:
- To develop a lightweight DL model for accurate and efficient recognition of HHIs using WiFi CSI.
- To introduce an attention mechanism that enhances feature representation for HHI recognition.
- To address the limitations of existing DL architectures in terms of accuracy and computational complexity.
Main Methods:
- Proposed HHI-AttentionNet, a dynamic and lightweight DL model for HHI recognition.
- Integrated an Antenna-Frame-Subcarrier Attention Mechanism (AFSAM) to improve feature selection and representation.
- Evaluated the model on a public CSI-based HHI dataset involving 40 pairs performing 13 different HHIs.
Main Results:
- HHI-AttentionNet achieved high performance metrics: 95.47% accuracy, 95.45% F1 score, 0.951% Cohen's Kappa, and 0.950% Matthews correlation coefficient.
- The model demonstrated superior performance compared to existing methods, outperforming the best by over 4% in accuracy.
- The attention mechanism effectively focused on significant features and reduced the impact of CSI signal complexity.
Conclusions:
- HHI-AttentionNet offers a significant advancement in WiFi-based HHI recognition.
- The lightweight design and enhanced accuracy make it suitable for practical, resource-limited applications.
- The proposed AFSAM is effective in improving the representational capability for complex human interactions.
