Publication

Abstract : Sign language recognition (SLR) systems are vital in bridging communication gaps and enhancing accessibility for deaf and hard-of-hearing individuals. This work presents a novel approach that combines spatial and temporal feature learning in a single model. It captures both the configurations of the hands and the patterns of gestures in sequence. Unlike traditional single-stream architectures, the hybrid model captures the complex motion interdependencies of signing and is attuned to subtle frame changes. To assess its performance, the model was evaluated on two benchmark datasets: Chinese Sign Language (CSL) and British Sign Language (BSL), both featuring 25 gesture categories. The model achieved notable recognition rates of 91.46% on CSL and 88.87% on BSL, illustrating strong adaptability across cultures. These findings assert the promise of such hybrid architectures for real-world sign-to-text systems and demonstrate the need for developing systems for human–computer interaction (HCI) that are more accurate and dependable.