Dynamic activity classification based on automatic adaptation of postural orientation.

Area of Science:

• Wearable technology
• Human activity recognition
• Biomedical engineering

Background:

• Activity classification relies on tri-axial accelerometer sensors, typically worn at fixed positions.
• Sensor tilt variations due to user posture changes (sitting, standing, lying) can significantly degrade performance.
• Users prefer flexible sensor placement, often deviating from recommended positions.

Purpose of the Study:

• To develop a dynamic activity classification system adaptable to user postural orientation.
• To enable flexible sensor placement on the torso without compromising accuracy.
• To overcome limitations of fixed-tilt activity detection methods.

Main Methods:

• Proposed a user-adapted activity classification method for tri-axial accelerometer sensors.
• Focused on tilt-free activities like walking and running to determine postural orientation.
• Developed an algorithm to dynamically modify reference tilt values for sensor axes (X, Y, Z) during walking.

Main Results:

• Achieved 88% activity classification accuracy in experimental validation.
• Demonstrated system robustness despite changes in sensor tilt angle.
• Successfully enabled classification of activities with user-preferred sensor placement.

Conclusions:

• The proposed dynamic system effectively adapts to user postural orientation for accurate activity classification.
• User-adapted methods enhance the practicality and reliability of wearable sensor-based activity recognition.
• This approach addresses key challenges in real-world application of human activity monitoring.