Deep vision-based real-time hand gesture recognition: a review.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Human-Computer Interaction

Background:

• Hand gesture recognition is crucial for human-computer interaction but faces challenges like background noise, motion blur, and processing delays.
• Deep learning approaches, including CNN, LSTM, and attention mechanisms, have emerged to address these limitations.

Purpose of the Study:

• To review and compare deep learning models (CNN, LSTM, attention mechanisms) for hand gesture recognition.
• To analyze evaluation metrics, datasets, and ablation studies used in deep learning-based gesture recognition.
• To identify research gaps and suggest future research directions.

Main Methods:

• Comparative analysis of Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), and attention mechanisms.
• Discussion of evaluation metrics (accuracy, efficiency), datasets (self-created, public), and ablation study methodologies.
• Synthesis of existing research on deep learning for hand gesture recognition.

Main Results:

• CNN enhances edge clarity, LSTM improves rotation accuracy, and attention mechanisms optimize response time in hand gesture recognition.
• Performance evaluation is critically dependent on the choice of metrics and dataset diversity.
• Ablation studies reveal insights into model components, preprocessing, and integration strategies.

Conclusions:

• Deep learning models offer significant improvements in hand gesture recognition accuracy and efficiency.
• Further research is needed to address accuracy, efficiency, application range, and environmental adaptation.
• Optimizing model performance requires careful consideration of evaluation metrics, datasets, and experimental design.