Real-World Gait Speed Estimation Using Wrist Sensor: A Personalized Approach.

Area of Science:

• Biomechanics
• Wearable technology
• Machine learning

Background:

• Gait speed is crucial for daily mobility assessment.
• Wrist-worn inertial sensors offer potential for gait speed measurement but face accuracy limitations.
• Global Navigation Satellite System (GNSS) provides accurate data but has high power demands and outdoor limitations.

Purpose of the Study:

• To develop a personalized gait speed estimation model using wrist-mounted sensors.
• To improve accuracy and reduce power consumption compared to existing methods.
• To enable accurate, real-time, low-power indoor/outdoor speed estimation.

Main Methods:

• Extracted gait features (cadence, etc.) from wrist accelerometer and barometer.
• Fused inertial sensor data with limited, sporadically sampled GNSS data.
• Employed online learning for personalized step length calibration.

Main Results:

• Achieved median RMSE of 0.05 m/s (walking) and 0.14 m/s (running).
• Personalized model demonstrated performance comparable to full GNSS.
• Reduced GNSS training data requirement by 50x compared to non-personalized methods.

Conclusions:

• The personalized model offers accurate and efficient gait speed estimation.
• Reduced GNSS reliance enhances battery life for wearable devices.
• The algorithm meets requirements for diverse real-world mobility monitoring applications.