Sign Language Recognition Using Graph Convolu...

Sign Language Recognition Using Graph Convolutional Networks Integrated with Attention and Residual Connections

Author Name : Dr. P. Vasuki, Madhupriya C, Karthikayan V, Mano Ranjan J

ABSTRACT Sign Language Recognition (SLR) enhances communication for individuals with hearing impairments by surpassing traditional interpreter-based methods. Most existing SLR systems rely on hand skeleton joint data to address occlusions and background noise, yet they often overlook the role of body motion and facial expressions in sign language. While multi-gesture-based SLR models exist, they struggle with real-time performance and accuracy. To overcome these limitations, we propose a two-stream multistage Graph Convolutional Network with Attention and Residual connections (GCAR), designed for improved spatial-temporal feature extraction. Our model processes joint skeleton data through two parallel streams, employing graph convolutions, deep learning layers, and a channel attention module. By merging these outputs, the system effectively captures structural displacements and short range dependencies. With only 0.69 million parameters, our method achieves high accuracy on multiple datasets, demonstrating superior performance and computational efficiency. This approach sets a new benchmark for sign language recognition research.