Augmented Movelet Method for Activity Classification Using Smartphone Gyroscope and Accelerometer Data

Area of Science:

• Biomedical Engineering
• Wearable Technology
• Human Activity Recognition

Background:

• Traditional surveys for physical activity measurement are subjective and prone to recall bias.
• Objective measurement of physical activity is crucial in biomedical research and clinical settings.
• Smartphones offer a potential for unobtrusive, objective physical activity monitoring.

Purpose of the Study:

• To explore the potential of smartphone accelerometer and gyroscope data for distinguishing between various physical activities.
• To develop a method for classifying activity types and quantifying classification uncertainty using smartphone sensor data.

Main Methods:

• Four participants performed a sequence of activities (walking, sitting, standing, ascending/descending stairs) with smartphones in front and back pockets.
• Activity was recorded via video and annotated to establish ground truth.
• A modified movelet method was used to classify activity types from sensor data.

Main Results:

• Smartphone sensor data can distinguish between different physical activities.
• The modified movelet method demonstrated potential for activity classification and uncertainty quantification.
• Results highlight both the promise and challenges of using smartphones for activity recognition in real-world settings.

Conclusions:

• Smartphones show promise for objective, unobtrusive physical activity measurement in naturalistic environments.
• Further research is needed to address challenges in analyzing noisy smartphone sensor data for reliable activity recognition.