Analytical Validation of Wrist-Worn Accelerometer-Based Step-Count Methods during Structured and Free-Living...

Area of Science:

• Biomedical Engineering
• Wearable Technology
• Activity Recognition

Background:

• Wrist-worn accelerometers offer passive, continuous, and remote monitoring of stepping behavior.
• Existing step-counting methods (peak detection, threshold crossing, frequency analysis) have unclear performance across diverse activities and speeds.
• Evaluating open-source algorithms for wrist-worn accelerometers is crucial for accurate activity monitoring.

Purpose of the Study:

• To evaluate the performance of four open-source step-counting methods using wrist-worn accelerometry data.
• To assess the impact of parameter modification on method performance.
• To compare method accuracy across structured locomotion and free-living activities.

Main Methods:

• Twenty-one participants wore wrist-worn accelerometers during laboratory-based structured locomotion and free-living activities.
• Criterion step counts were obtained via motion capture and a secondary step-counting device.
• Four open-source algorithms were applied to accelerometer data, with and without a locomotion classifier.

Main Results:

• Single-parameter methods (peak detection, threshold crossing) showed lowest bias in structured locomotion.
• Three methods overestimated steps during slow walking and underestimated during fast walking.
• Frequency analysis method had the lowest percent error during free-living activities.
• A locomotion classifier reduced error for two methods across both activity types.

Conclusions:

• A trade-off exists between step-counting accuracy in structured walking versus free-living activities.
• Current open-source methods show variable performance depending on activity type and speed.
• Development of context-aware algorithms is recommended for improved accuracy in real-world step counting.