Multiple Classification of Gait Using Time-Frequency Representations and Deep Convolutional Neural Networks.

Area of Science:

• Biomechanics
• Computational neuroscience
• Medical imaging analysis

Background:

• Human gait analysis is crucial for health assessment.
• Previous gait classification methods struggle with multi-classification and subtle differences.
• Need for advanced techniques to differentiate gait characteristics without visual cues.

Purpose of the Study:

• To propose a novel spectrographic approach for multi-classification of human gait.
• To accurately classify gait with no visually discernible differences.
• To leverage deep convolutional neural networks (CNNs) for gait analysis.

Main Methods:

• Recruited 69 participants: semi-professional athletes, ordinary individuals, and those with subtle foot deformities.
• Collected 3-axis acceleration and angular velocity data using inertial measurement units (IMUs) on the lower body.
• Applied time-frequency analysis to create gait spectrograms for training CNN classifiers.

Main Results:

• The spectrographic approach enabled complete classification of three distinct gait groups.
• Classification accuracy was high even without complex feature engineering.
• Foot, shank, and thigh spectrograms were sufficient for reliable multi-classification.

Conclusions:

• The spectrographic approach combined with CNNs offers a reliable method for multi-class gait classification.
• This technique can differentiate subtle gait variations previously indistinguishable.
• This study pioneers the use of spectrographic analysis in gait classification for health monitoring.