Gait Environment Recognition Using Biomechanical and Physiological Signals with Feed-Forward Neural Network: A Pilot...

Area of Science:

• Biomechanics and Robotics
• Wearable Sensor Technology
• Machine Learning for Human Locomotion

Background:

• Human locomotion, or gait, is essential for daily activities and occurs in varied environments.
• Recognizing environmental changes during walking is critical for fall prevention and controlling assistive wearable robots.
• Current methods for gait environment classification often lack real-time adaptability.

Purpose of the Study:

• To develop and validate a novel system for classifying different gait environments using wearable sensor data.
• To assess the efficacy of a feed-forward neural network (FFNN) in distinguishing between level ground, ramps, and stairs.
• To establish a foundation for advanced gait analysis and exoskeleton robot control.

Main Methods:

• Collected synchronized gait data from five participants using inertial measurement units, galvanic skin response sensors, and smart insoles across level ground, ramps, and stairs.
• Preprocessed 47,033 data samples through time synchronization and filtering, followed by environment-specific labeling.
• Trained a feed-forward neural network (FFNN) model with a single hidden layer to classify the gait environments.

Main Results:

• The FFNN model achieved a high classification accuracy of 98% for identifying gait environments.
• The highest accuracy was observed during walking on level ground.
• The study demonstrated the effectiveness of combining data from multiple wearable sensors for gait environment recognition.

Conclusions:

• Wearable sensor technology combined with machine learning effectively classifies diverse gait environments.
• This approach shows significant promise for enhancing the safety and functionality of wearable robotic systems.
• The findings provide crucial baseline data for future research in exoskeleton control and sophisticated gait analysis.