Classification of Gait Type Based on Deep Learning Using Various Sensors with Smart Insole

Area of Science:

• Biomedical Engineering
• Machine Learning
• Human Movement Analysis

Background:

• Gait analysis is crucial for diagnosing neurological disorders and assessing rehabilitation progress.
• Traditional gait analysis methods often require specialized equipment and controlled laboratory environments.
• Developing unobtrusive, accurate, and accessible gait monitoring systems is a significant challenge.

Purpose of the Study:

• To propose and validate a novel gait type classification method using a smart insole equipped with multiple sensor arrays.
• To leverage deep learning, specifically deep convolution neural networks (DCNNs), for extracting gait features from sensor data.
• To achieve high classification accuracy across various common and challenging gait types.

Main Methods:

• A smart insole integrated with pressure, acceleration, and gyro sensor arrays was developed to collect gait data.
• Continuous gait cycle data were segmented into unit steps, followed by noise removal and data normalization pre-processing.
• Independent DCNNs were employed to extract feature maps from each sensor array's data.
• A fully connected network combined these feature maps for the final gait type classification.

Main Results:

• The proposed deep learning method demonstrated high performance in classifying seven distinct gait types: walking, fast walking, running, stair climbing, stair descending, hill climbing, and hill descending.
• The classification accuracy consistently exceeded 90% across all tested gait types.
• The multi-sensor fusion approach effectively captured complex gait patterns.

Conclusions:

• The smart insole-based deep learning method offers a highly accurate and robust solution for automatic gait type classification.
• This technology has the potential for real-world applications in healthcare, sports science, and human-computer interaction.
• The DCNN-driven feature extraction and fusion strategy proved effective for complex human locomotion analysis.