Classification of human physical activity based on raw accelerometry data via spherical coordinate transformation.

Area of Science:

• Biomedical Engineering
• Human Movement Science
• Wearable Technology

Background:

• Human health is closely linked to lifestyle and physical activity levels.
• Accurate characterization of daily human activity is crucial for health monitoring.
• Wearable accelerometers provide precise, high-frequency 3D time-series data of body acceleration.

Purpose of the Study:

• To develop a novel procedure for classifying physical activity using raw accelerometry signals.
• To explore the effectiveness of a spherical data representation for activity classification.
• To classify four distinct activity types: resting, sitting, standing, and lower body activities.

Main Methods:

• A novel classification procedure based on the spherical representation of raw accelerometry data.
• Feature extraction included spherical coordinates summary statistics, moving averages, radius variance, and spherical variance.
• A decision tree classifier was employed for activity classification.
• The method was tested on data from 47 elderly individuals in laboratory settings.

Main Results:

• The proposed method achieved over 90% classification accuracy using subject-specific data.
• Group data analysis yielded an 84% classification accuracy.
• The angular component of the signal, particularly spherical variance, was the main contributor to accuracy.
• Spherical variance demonstrated invariance to accelerometer location shifts.

Conclusions:

• The spherical representation of accelerometry data offers a promising approach for accurate physical activity classification.
• Spherical variance is a novel and effective feature for analyzing raw accelerometry data, outperforming other angular measures.
• This method has significant potential for improving health monitoring through objective physical activity assessment.