Isolated Sign Language Recognition from RGB Video

Abstract

Sign Language Recognition (SLR) is a key task for breaking communication barriers for the deaf community. This thesis addresses major challenges in Isolated SLR (ISLR): data scarcity, complex motion modeling, and reliance on specialized sensors. We propose novel methodologies using transfer learning and attention mechanisms, enabling robust ISLR systems that achieve high accuracy and data efficiency using only ubiquitous RGB camera inputs.