An integrated mediapipe-optimized GRU model for Indian sign language recognition.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Sign language recognition faces challenges like hand gesture tracking, occlusion, and high computational costs.
• Deep learning models struggle with long-term sequential data, leading to poor information processing and learning efficiency.

Purpose of the Study:

• To propose an integrated MediaPipe-optimized gated recurrent unit (MOPGRU) model for enhanced Indian sign language recognition.
• To address the limitations of existing sequential models in processing long-term data and capturing relevant information.

Main Methods:

• Developed the MOPGRU model by modifying the update gate of standard GRU cells to discard redundant past information.
• Incorporated feedback from the reset gate to focus on present input, enhancing attention.
• Replaced hyperbolic tangent activation with exponential linear unit and SoftMax with Softsign in the output layer.

Main Results:

• The MOPGRU model demonstrated superior prediction accuracy compared to other sequential models.
• Achieved higher learning efficiency and improved information processing capabilities.
• Exhibited faster convergence rates in sign language recognition tasks.

Conclusions:

• The proposed MOPGRU model effectively overcomes challenges in sign language recognition.
• Offers significant improvements in accuracy, efficiency, and processing capabilities for Indian sign language.
• Represents a promising advancement in deep learning for sign language recognition systems.