A Comprehensive Review of Sign Language Recognition Systems: Methods, Datasets, Challenges, and Future Directions

Abstract

Sign Language Recognition (SLR)has developed as an essential research area that has the purpose of allowing  equal opportunities for communication between deaf or hard-of-hearing individuals and the hearing population.  Recent associated advances in deep learning, multimodal sensing, and transformer-based architectures have driven  SLR systems toward higher accuracy, robustness, and real-time performance across diverse linguistic contexts.  This review presents a comprehensive and technically grounded analysis of SLR systems, encompassing isolated,  continuous, multilingual, multimodal, and translation-oriented research. By systematically synthesizing  contributions from vision-based systems, inertial sensing, electromyography, radar, wearable devices, and  emerging LLM-driven translation frameworks, the survey covers methodological progress and identifies current  limitations. The significant developments include the integration of spatio-temporal learning, attention-enhanced  CNN–RNN hybrids, pose-driven transformer models, radar-based sensing, MediaPipe-driven real-time systems,  and multimodal conformer architectures. Furthermore, recent work incorporating LLMs into cross-lingual sign  translation,, illustrates a drift toward semantically aware and context-sensitive pipelines in SLR. Besides the  methodological review, this paper covers an extensive review of major datasets, discusses challenges regarding  signer variability, occlusion, continuous-gesture transitions, domain gaps, and linguistic complexities, and  provides an outline for future directions on scalable, low-resource, and generalizable SLR technologies. This work  consolidates the present state of SLR research and provides a sound basis for further steps in next-generation  intelligent sign language communication systems.

DOI: https://doi.org/10.24321/3051.424X.202603