A comprehensive IMU dataset for evaluating sensor layouts in human activity and intensity recognition

Area of Science:

• Biomedical Engineering
• Computer Science
• Wearable Technology

Background:

• Human activity recognition (HAR) using wearable sensors is crucial for health monitoring and smart environments.
• Optimizing sensor configuration is key to balancing performance, usability, and comfort.
• Existing datasets often lack comprehensive full-body coverage for systematic sensor placement evaluation.

Purpose of the Study:

• To introduce a comprehensive dataset for evaluating sensor configurations in HAR.
• To facilitate systematic research into optimal sensor placement strategies.
• To support the development of practical and generalizable HAR systems.

Main Methods:

• Collected data from 30 participants performing 12 daily activities.
• Utilized 17 inertial measurement units (IMUs) across the entire body, capturing tri-axial acceleration and angular velocity at 60 Hz.
• Included detailed anthropometric metadata, activity/effort annotations, and processing scripts for feature extraction and model training.

Main Results:

• Demonstrated dataset usability through benchmark experiments with machine learning and deep learning models.
• Showcased performance across various temporal windows and sensor subsets.
• Validated the dataset's utility for evaluating sensor layout strategies.

Conclusions:

• The presented dataset enables systematic evaluation of sensor configurations for HAR.
• It supports the development of more practical and generalizable HAR systems.
• This resource advances research in wearable sensor-based activity recognition.