tinyRadar for Fitness: A Contactless Framework for Edge Computing.

Area of Science:

• Biomedical Engineering
• Wearable Technology
• Signal Processing

Background:

• Conventional fitness trackers face challenges like skin discomfort, inaccurate readings, and privacy risks due to prolonged wear and data transmission.
• The shift towards personalized healthcare necessitates innovative solutions for continuous, reliable health monitoring.

Purpose of the Study:

• To introduce tinyRadar, an on-edge millimeter wave (mmWave) radar-based fitness tracker designed to overcome the limitations of current wearable devices.
• To develop a compact, privacy-preserving fitness tracker suitable for smart home environments.

Main Methods:

• Utilized a Texas Instruments IWR1843 mmWave radar board for exercise recognition and repetition counting.
• Implemented signal processing and a Convolutional Neural Network (CNN) on-board for real-time analysis.
• Interfaced the radar board with ESP32 for data transfer to smartphones via Bluetooth Low Energy (BLE).

Main Results:

• Achieved 96% average accuracy in real-time repetition counting.
• Demonstrated 97% subject-independent classification accuracy for exercise types.
• The on-board CNN model exhibited low memory utilization (11.36 KB total, 1.46 KB for parameters).

Conclusions:

• tinyRadar effectively addresses discomfort and privacy concerns associated with traditional fitness trackers.
• The mmWave radar-based approach provides accurate and efficient exercise tracking in a small form factor.
• This technology is well-suited for personalized, real-time fitness monitoring within a smart home setting.