Human Gait Activity Recognition Machine Learning Methods.

Area of Science:

• Biomedical Engineering
• Wearable Technology
• Human Motion Analysis

Background:

• Human gait activity recognition is crucial for monitoring patients with gait disorders and evaluating treatment efficacy.
• Existing methods may lack robustness or personalization for diverse patient needs.

Purpose of the Study:

• To develop a robust, wearable system for acquiring and classifying human gait motion data.
• To identify the most effective machine learning algorithm for gait activity classification and risk factor identification.
• To enhance the quality of life for individuals with gait disorders through personalized monitoring.

Main Methods:

• Gait motion data collected using accelerometers and gyroscopes on lower limbs.
• Leg muscle activity measured via strain gauge sensors.
• Machine learning algorithms, including attention-based convolutional and recurrent neural networks, were employed for classification.

Main Results:

• The combined attention-based convolutional and recurrent neural networks achieved high classification performance: 98.9% accuracy, 96.8% precision, 97.8% sensitivity, 99.1% specificity, and 97.3% F1-score.
• The system successfully detected freezing gait episodes in a Parkinson's disease patient, demonstrating robustness.
• The algorithm proved capable of complete personalization for gait event classification.

Conclusions:

• A feasible and robust gait event classification method has been developed.
• The system offers potential for personalized monitoring and management of gait disorders.
• This technology can significantly improve patient quality of life through enhanced disease tracking and treatment evaluation.