Authors: Prerna Charis J
Abstract: Sign Language Recognition (SLR) has emerged as an important research area at the intersection of computer vision and deep learning, and human and machine interaction with an objective of enabling effective communication between deaf and hearing communities. Recent advances in deep learning have improved the performance of vision-based Sign Language Recognition systems, particularly by using hybrid architectures that combine spatial features extraction and temporal sequence modelling. The goal of this review is to provide a overview of the recent developments in hybrid Vision-based Sign Language Recognition and to examine the advantages, limitation and practical deployment challenges of the current approaches. This paper provides a systematic review of the literature, the surveyed methods broadly classified into CNN-LSTM architectures, Transformer-based models and multimodal integrated frameworks which integrates visual and skeletal information. This review further investigates critical challenges affecting the deployment in real-world scenarios which includes domain shift, data scarcity, co-articulation, sign ambiguity and computational constrain. We will also discuss about emerging research direction such as self-supervised learning, cross-linguistic transfer learning, generative domain adaptation, multimodal bio signal integration, and community-centered dataset development. This survey also highlights the significant progress achieved in continuous sign language recognition while identifying the remaining technical and practical barriers that must be removed to develop robust, scalable, and user-independent SLR systems capable of operating in real-world environments.
