Classifying running speed conditions using a single wearable sensor: Optimal segmentation and feature extraction...

Area of Science:

• Biomechanics
• Wearable Technology
• Signal Processing

Background:

• Accelerometers are used for classifying running patterns.
• Classification accuracy and computational load depend on signal segmentation and feature extraction methods.
• Stride-based segmentation requires gait event identification, unlike window-based segmentation.

Purpose of the Study:

• To examine how different segmentation and feature extraction methods influence the accuracy and computational load of classifying running conditions.
• To compare discrete versus advanced features.
• To compare stride-based versus window-based segmentation and small versus large segment sizes.

Main Methods:

• Forty-four runners ran at preferred and faster speeds, with 3D accelerations recorded by a lower back accelerometer.
• Accelerometer signals were segmented into single/five strides and small/large windows.
• Discrete points and advanced features were extracted from segments.
• Feature sets were used to classify speed conditions, recording accuracy and computational load.

Main Results:

• The five-stride and large-window advanced feature sets achieved the highest classification accuracy (97.49% and 97.23%, respectively).
• The large-window advanced feature set exhibited a lower computational load compared to stride-based feature sets.
• Advanced features combined with large window sizes offered the best balance of accuracy and computational efficiency.

Conclusions:

• Utilizing advanced features with large, overlapping window sizes optimizes both classification accuracy and computational load for running condition analysis.
• This approach offers a practical and efficient method for real-time running pattern classification using wearable sensors.