Gait Phase Recognition Using Deep Convolutional Neural Network with Inertial Measurement Units.

Area of Science:

• Robotics
• Biomedical Engineering
• Machine Learning

Background:

• Accurate gait phase recognition is essential for developing effective assistance-as-needed robotic devices like exoskeletons.
• Wearable sensors, specifically inertial measurement units (IMUs), offer a promising avenue for capturing user kinematics.
• Existing methods require precise identification of gait phases for seamless exoskeleton control and user support.

Purpose of the Study:

• To develop and evaluate a specialized deep convolutional neural network (DCNN) for precise gait phase recognition.
• To utilize inertial measurement unit (IMU) data, treated as sequential 'images', for gait phase classification.
• To achieve high accuracy in identifying five distinct phases within a human gait cycle.

Main Methods:

• A deep convolutional neural network (DCNN) architecture was designed and implemented.
• User kinematic data from IMUs was processed and formatted for DCNN input.
• Foot switch information was used for classifying and validating the recognized gait phases.
• The DCNN model was evaluated through offline testing for gait phase recognition accuracy.

Main Results:

• The specialized DCNN achieved an overall accuracy of approximately 97% in gait phase recognition.
• The highest accuracy was observed during the swing phase of the gait cycle.
• The lowest accuracy was recorded during the terminal stance phase.

Conclusions:

• Deep convolutional neural networks (DCNNs) are effective for gait phase recognition using IMU data.
• The proposed DCNN model demonstrates high potential for real-time application in exoskeleton control systems.
• Further refinement may improve accuracy in challenging gait phases like terminal stance.