Enhancing Multi-User Activity Recognition in an Indoor Environment with Augmented Wi-Fi Channel State Information and...

Area of Science:

• Computer Science
• Signal Processing
• Artificial Intelligence

Background:

• Human Activity Recognition (HAR) is vital for monitoring behavior using sensor data, but traditional methods face limitations.
• Wearable and vision-based HAR systems raise privacy concerns and struggle in dynamic environments.
• Device-free HAR using Wi-Fi Channel State Information (CSI) offers a privacy-preserving alternative, especially for multi-user scenarios.

Purpose of the Study:

• To address data scarcity and generalisability challenges in multi-user device-free HAR.
• To propose a hybrid deep learning model integrating CNN and Transformer architectures.
• To improve HAR performance in complex environments with limited labeled data.

Main Methods:

• Implemented a hybrid deep learning approach combining Convolutional Neural Networks (CNN) and Transformer models.
• Utilized a random transformation technique for targeted data augmentation of real CSI data.
• Employed hybrid feature extraction including statistical, spectral, and entropy-based measures.

Main Results:

• The proposed model demonstrated superior performance compared to baseline methods in both single-user and multi-user contexts.
• Combining real and augmented CSI data significantly enhanced model generalisation.
• Effectively addressed challenges of data scarcity and class imbalance in multi-user HAR.

Conclusions:

• The hybrid deep learning model offers a robust solution for device-free Human Activity Recognition.
• Data augmentation and hybrid feature extraction are crucial for improving HAR in data-scarce, multi-user environments.
• This approach advances privacy-preserving human behavior monitoring in smart environments.