A Fusion Recognition Method Based on Multifeature Hidden Markov Model for Dynamic Hand Gesture.

Area of Science:

• Robotics
• Human-Computer Interaction
• Machine Learning

Background:

• Robot teleoperation requires intuitive control methods.
• Recognizing dynamic hand gestures is crucial for effective operator instructions.
• Existing methods may lack robustness in complex gesture recognition.

Purpose of the Study:

• To propose a fusion method for dynamic hand gesture recognition.
• To enhance operator instruction interpretation in robot teleoperation.
• To improve the accuracy and reliability of gesture recognition systems.

Main Methods:

• A fusion method combining multiple features and Hidden Markov Models (HMM).
• Gesture segmentation based on hand velocity from continuous data.
• Feature extraction including palm posture, finger bending/opening angles, and trajectory.
• Weighted probability fusion model for HMM classifiers.

Main Results:

• High recognition rates achieved: 90.63% on the LM-Gesture3D dataset.
• Excellent performance demonstrated: 93.3% on the Letter-gesture dataset.
• Validation using Leap Motion (LM) sensor data.

Conclusions:

• The proposed multi-feature fusion HMM method is effective for dynamic hand gesture recognition.
• This approach significantly improves gesture recognition accuracy in robot teleoperation.
• The method offers a robust solution for interpreting operator commands via hand gestures.