EHUNAM, a WiFi CSI-based dataset for human and machine sensing.

Area of Science:

• Computer Science
• Electrical Engineering
• Signal Processing

Background:

• WiFi Sensing (WS) applications require high-quality, diverse datasets for cross-domain capability.
• Existing datasets may lack the variability needed for advanced WS tasks.

Purpose of the Study:

• Introduce EHUNAM, a comprehensive Channel State Information (CSI) dataset for WiFi Sensing.
• Support applications such as people counting (PC), human activity recognition (HAR), and machine activity recognition (MAR).

Main Methods:

• Collected CSI data over 23 days in eight diverse environments, including industrial settings.
• Included data from 21 people and nine machines, with simultaneous activities.
• Utilized varied equipment configurations and scenarios for dataset representativeness.

Main Results:

• Validated EHUNAM using a Convolutional Neural Network (CNN) for PC, HAR, and MAR.
• Achieved over 90% accuracy in most multiclass and multilabel classification tasks.
• Demonstrated the dataset's robustness for a broad spectrum of real-world WS scenarios.

Conclusions:

• EHUNAM is a versatile and robust dataset for advancing WiFi Sensing applications.
• The dataset's quality and variability enhance cross-domain capabilities in WS.
• EHUNAM supports recognition of human and machine activities, including home appliances and industrial machines.