High Accuracy WiFi-Based Human Activity Classification System with Time-Frequency Diagram CNN Method for Different...

Area of Science:

• Engineering
• Computer Science
• Gerontology

Background:

• Falls pose a significant health risk to the elderly, necessitating effective fall detection systems.
• Traditional fall detection methods using wearables or image recognition have limitations like user-unfriendliness and privacy concerns.
• Existing WiFi-based systems struggle with accuracy due to environmental variations between training and application sites.

Purpose of the Study:

• To develop a robust WiFi-based fall detection system that overcomes the accuracy limitations of current methods.
• To leverage Channel State Information (CSI) and Convolutional Neural Networks (CNNs) for enhanced fall detection.
• To validate the proposed system's performance in diverse environmental conditions.

Main Methods:

• Utilized Channel State Information (CSI) extracted from WiFi signals to capture human motion.
• Employed Convolutional Neural Networks (CNNs) for feature extraction and action classification from CSI data.
• Developed and tested a prototype system to evaluate the proposed fall detection method.

Main Results:

• The proposed system achieved an average accuracy of 93.2% when tested in the same location as training.
• The system demonstrated strong performance with an average accuracy of 90.3% in a different location than training.
• These results indicate the system's effectiveness in generalizing to new environments.

Conclusions:

• The developed WiFi-based fall detection system using CSI and CNNs offers a promising solution for elderly fall prevention.
• The method shows high accuracy and adaptability to different environments, addressing key limitations of existing systems.
• This technology has the potential to improve safety and independence for older adults.