Domain Adaptation with Contrastive Simultaneous Multi-Loss Training for Hand Gesture Recognition.

Area of Science:

• Computer Vision
• Machine Learning
• Human-Robot Interaction

Background:

• Hand gesture recognition is vital for human-robot interaction, especially in industrial settings.
• Accurate hand segmentation is difficult in noisy, unstructured industrial environments.
• Current methods rely on extensive preprocessing before deep learning classification.

Purpose of the Study:

• To develop a more robust and generalizable hand gesture classification model.
• To address challenges in hand segmentation within industrial collaborative scenarios.
• To improve hand gesture recognition performance using domain adaptation.

Main Methods:

• Proposed a novel domain adaptation approach utilizing multi-loss training and contrastive learning.
• Tested the model on an unrelated dataset with different users to assess generalizability.
• Employed simultaneous multi-loss functions combined with contrastive learning techniques.

Main Results:

• The proposed approach demonstrated superior performance in hand gesture recognition.
• Achieved better results compared to conventional methods in challenging conditions.
• Validated the effectiveness of contrastive learning within multi-loss functions.

Conclusions:

• The developed method offers a more effective solution for hand gesture recognition in difficult environments.
• Domain adaptation with multi-loss and contrastive learning enhances model robustness and generalizability.
• This approach is particularly promising for industrial collaborative robotics.