Recognition of Upper Limb Action Intention Based on IMU.

Area of Science:

• Biomedical Engineering
• Robotics
• Human-Computer Interaction

Background:

• Controlling prosthetic hands requires accurate recognition of user's upper limb motion intentions.
• Wearable inertial sensors offer a practical solution for capturing limb movement data due to their size, cost, and low interference.
• Existing methods may face challenges in real-time, intention-driven prosthetic control.

Purpose of the Study:

• To develop and validate novel algorithms for classifying upper limb actions for prosthetic hand control.
• To improve the accuracy and reduce the response time of prosthetic hand action recognition.
• To enable intuitive prosthetic hand grasping intention recognition based on limb movement.

Main Methods:

• Utilized wearable inertial sensors to collect upper limb angle and angular velocity data.
• Proposed a Dynamic Time Warping (DTW) algorithm based on motion units for action classification (putting on socks, shoes, tying shoelaces).
• Developed a Generalized Regression Neural Network (GRNN) model with 10-fold cross-validation for grasping intention recognition, using static limb state as a control signal.

Main Results:

• The DTW-based motion unit recognition model achieved a high recognition rate of 99.46% with an average running time of 8.027 ms.
• The GRNN model with 10-fold cross-validation demonstrated a high accuracy rate of 98.28% for grasping intention recognition.
• Experimental validation confirmed the feasibility and practicality of both algorithms in real-time prosthetic hand action reproduction.

Conclusions:

• The proposed DTW algorithm effectively classifies upper limb actions for prosthetic control with superior accuracy and speed.
• The GRNN model provides a robust method for recognizing grasping intentions, enhancing prosthetic hand responsiveness.
• These algorithms represent a significant advancement in creating more intuitive and functional prosthetic limb systems.