Gait Segmentation of Data Collected by Instrumented Shoes Using a Recurrent Neural Network Classifier

Area of Science:

• Biomechanics
• Machine Learning
• Wearable Technology

Background:

• Accurate gait analysis is crucial for understanding human locomotion and diagnosing movement disorders.
• Instrumented footwear offers a promising method for unobtrusive gait data collection.
• Automated segmentation of gait cycles is essential for efficient and scalable analysis.

Purpose of the Study:

• To develop and validate a Recurrent Neural Network (RNN) classifier for segmenting walking data from instrumented footwear.
• To assess the model's accuracy in identifying temporal gait characteristics, specifically heel strikes and toe-offs.
• To demonstrate the reliability of the proposed algorithm for gait analysis.

Main Methods:

• Utilized a Recurrent Neural Network (RNN) classifier model.
• Input signals included data from 3 piezoresistive sensors, a 3-axis accelerometer, and Euler angles.
• The model was trained and tested on a dataset comprising 4198 steps from 28 adult participants.

Main Results:

• Achieved mean errors of -5.9 ± 37.1 milliseconds for heel strikes.
• Achieved mean errors of 11.4 ± 47.4 milliseconds for toe-offs.
• Demonstrated high reliability in segmenting gait recordings.

Conclusions:

• The developed RNN classifier reliably segments gait data from instrumented footwear.
• The algorithm's accuracy in identifying gait events allows for precise estimation of temporal gait parameters.
• This approach offers a robust solution for automated gait analysis in various applications.