A Hybrid Kinematic GCN and Elastic Transformer for Efficient Indian Sign Language Recognition

A hybrid framework for Indian Sign Language recognition combines a Kinematically-Conditioned Adaptive GCN with an Elastic Transformer to model bimanual geometry and irregular signing rhythms. The architecture utilizes entropy-driven SVD sampling for keyframe extraction and Continuous Dual Positional Encoding to maintain temporal consistency across variable signing velocities.