Multiple Participants' Discrete Activity Recognition in a Well-Controlled Environment Using Universal Software Radio...

Area of Science:

• Wireless sensing technology
• Human activity recognition
• Machine learning applications in healthcare

Background:

• Privacy-preserving health monitoring is crucial for smart homes and care centers.
• Existing systems struggle with multi-subject activity recognition.
• Need for advanced sensing for concurrent human activity monitoring.

Purpose of the Study:

• To develop a smart wireless sensing system for multi-subject human activity monitoring.
• To integrate radio-frequency sensing with ensemble machine learning for enhanced recognition.
• To simultaneously identify occupancy count and performed activities.

Main Methods:

• Utilized radio-frequency sensing operating at 3.75 GHz.
• Collected Channel State Information (CSI) amplitudes from 51 subcarriers.
• Employed an ensemble machine learning model trained on CSI data for activity recognition.
• Experiment involved up to four subjects performing sixteen daily living activities.

Main Results:

• Achieved high average accuracy of up to 98% for distinguishing multi-subject activities.
• Successfully merged subject count and activity recognition.
• Demonstrated effective capture of alterations from concurrent multi-subject motions.

Conclusions:

• The proposed system offers a viable solution for future health activity monitoring demands.
• Ensemble machine learning with CSI effectively recognizes complex multi-subject activities.
• Wireless sensing provides a privacy-preserving approach for well-being tracking.