Robust Stride Segmentation of Inertial Signals Based on Local Cyclicity Estimation.

Area of Science:

• Biomechanics
• Signal Processing
• Wearable Technology

Background:

• Gait analysis using inertial measurement units (IMUs) is crucial for understanding human movement.
• Accurate stride segmentation and gait event detection are fundamental challenges in processing IMU data.
• Existing methods often require specific sensor types, templates, or supervised learning.

Purpose of the Study:

• To introduce a novel, universal, and unsupervised approach for stride segmentation, gait sequence extraction, and gait event detection.
• To develop a method that is independent of sensor type and does not rely on predefined templates.
• To evaluate the performance of the proposed approach on diverse datasets and conditions.

Main Methods:

• Combining local cyclicity estimators and multiple sensor channels from IMUs.
• Utilizing a priori knowledge of gait event fiducial points.
• Implementing an unsupervised learning framework for gait analysis.

Main Results:

• Achieved high robustness and efficiency with an F-measure of approximately 98%.
• Demonstrated high accuracy with a mean absolute error (MAE) within a one-sample range.
• Validated performance across two distinct datasets (FRIgait and FAU eGait) including healthy, geriatric, and Parkinson's disease patients.
• Showcased effectiveness in both controlled and uncontrolled environments.

Conclusions:

• The proposed approach is a universal, template-free, and unsupervised solution for gait event detection from inertial signals.
• The method exhibits significant potential for widespread application in movement analysis procedures and algorithms.
• High accuracy and efficiency make it suitable for real-world applications in clinical and research settings.