Real-Time Hand Gesture Monitoring Model Based on MediaPipe's Registerable System.

Area of Science:

• Computer Vision
• Machine Learning
• Human-Computer Interaction

Background:

• Hand gesture recognition is crucial for interaction but current methods lack flexibility.
• Existing systems struggle with the diversity and variability of real-world hand gestures.

Purpose of the Study:

• To propose a novel registerable hand gesture recognition approach.
• To enable personalized gesture recognition by allowing users to register new gestures.

Main Methods:

• Developed a registerable gesture dataset (RGDS).
• Proposed a Triple Loss-based method for learning gesture differences and clustering.
• Introduced a normalization technique and a FingerComb block for feature enhancement.
• Improved ResNet and proposed FingerNet for registerable single-hand gesture recognition.

Main Results:

• The proposed FingerNet model demonstrated strong performance on the RGDS dataset.
• The system successfully achieved registerable hand gesture recognition.
• The approach effectively clusters and identifies newly added gestures.

Conclusions:

• The developed system offers a flexible and personalized solution for hand gesture recognition.
• This method addresses the limitations of fixed hand gesture recognition systems.
• The registerable nature allows for adaptable and user-specific gesture interaction.