System-Identification-Based Activity Recognition Algorithms With Inertial Sensors.

Area of Science:

• Biomedical Engineering
• Wearable Technology
• Human Activity Recognition

Background:

• Accurate human activity recognition is crucial for health monitoring, especially for individuals with conditions like Parkinson's disease (PD).
• Traditional methods often require multiple sensors or complex setups, limiting practical application.
• A single, unobtrusive sensor offers a promising avenue for continuous and accessible monitoring.

Purpose of the Study:

• To develop and validate a robust activity recognition system using a single wearable inertial measurement sensor.
• To identify ten distinct daily activities, including lying, standing, sitting, bending, and walking.
• To assess the system's efficacy in both clinical and remote home monitoring settings for Parkinson's disease patients.

Main Methods:

• Utilized a single chest-mounted inertial measurement sensor for data acquisition.
• Employed a transfer function identification approach, determining input-output signals based on activity-specific sensor signal norms.
• Applied Wiener filter with auto-correlation and cross-correlation for transfer function identification using training data.
• Real-time activity recognition achieved by comparing input-output errors across identified transfer functions.

Main Results:

• The developed system demonstrated an average accuracy exceeding 90% for identifying ten distinct activities.
• Performance was validated using data from Parkinson's disease subjects in both clinical and home environments.
• The system proved effective in real-time activity recognition.

Conclusions:

• A single wearable sensor system can achieve high accuracy for human activity recognition.
• This technology holds significant potential for monitoring Parkinson's disease patients' activity levels and fall risk.
• The system facilitates objective assessment of postural instability and real-time identification of high-risk activities.