Walking Speed Classification from Marker-Free Video Images in Two-Dimension Using Optimum Data and a Deep Learning...

Area of Science:

• Biomechanics
• Artificial Intelligence
• Medical Imaging

Background:

• Walking speed is a crucial indicator of functional mobility, utilized by clinicians and caregivers for patient assessment.
• Traditional gait monitoring methods can be enhanced by artificial intelligence (AI) for predicting physical outcomes and accidents.
• AI-driven gait analysis extracts body measurements from video to classify walking speed, but data optimization is key.

Purpose of the Study:

• To determine the optimal combination of ratio-based body measurements for accurate walking speed classification.
• To investigate how different measurement combinations impact deep learning model performance.
• To enhance the predictive power of AI models for gait pattern analysis.

Main Methods:

• Utilized marker-free, 2D video imaging to extract five ratio-based body measurements.
• Evaluated various combinations of these measurements through correlation analysis.
• Applied a bidirectional long short-term memory (BiLSTM) deep learning model for walking speed classification.

Main Results:

• Identified a specific combination of three ratio-based body measurements as optimal for gait analysis.
• Achieved classification accuracies exceeding 92% for predicting walking speed patterns.
• Demonstrated that optimized data input significantly improves deep learning model performance.

Conclusions:

• A reduced set of key body measurements can effectively predict walking speed with high accuracy.
• Deep learning models, particularly BiLSTM, are highly effective for AI-based gait analysis.
• Optimized data selection is critical for developing robust and efficient AI systems in healthcare.