A Critical Study of Recent Deep Learning-Based Continuous Sign Language Recognition

With the growing global emphasis on accessibility and inclusion for deaf and hard-of-hearing individuals, research in Sign Language Recognition (SLR) has gained significant momentum. Sign languages have unique grammar and syntax and rely on manual communication to convey meaning, which presents specific challenges for automated recognition systems. Unlike isolated sign language recognition, continuous sign language recognition (CSLR) must accurately interpret sequences of gestures without clear boundaries between signs. This requires advanced techniques for both segmentation and recognition. This paper presents a comprehensive review of Continuous Sign Language Recognition (CSLR), focusing specifically on deep learning (DL) techniques. It addresses key challenges, such as movement epenthesis (ME)—the transitional movements between signs. Implicit models, including Hidden Markov Models (HMMs) and Connectionist Temporal Classification (CTC), have demonstrated superior performance compared to traditional methods. The researchers reviewed 32 studies published by major publishers (IEEE, Elsevier, and Springer) and examined 12 benchmark datasets related to CSLR. This examination included an overview of their linguistic scope, recording setups, vocabulary size, and participant diversity. Additionally, the performance metrics and methods used in these studies were thoroughly analyzed, followed by a discussion of the results and their broader implications. The review highlights several limitations in existing research, underlining the need for ongoing innovation within the CSLR domain. The insights gained from this review not only enhance the understanding of Sign Language Recognition but also provide a foundation for future research aimed at tackling persistent challenges in this evolving field.