A Detailed Analysis of Recent Advances in Automatic Sign Language Recognition

EasyChair Preprint 14820

Abstract

This survey paper reviews the advancements in sign language recognition (SLR) and sign language translation (SLT) technologies. Both congenital and acquired Deaf and Hard of Hearing (DHH) people utilize sign language, a unique visual language that combines manual and nonmanual aspects for efficient communication. This paper explores various methods and models developed to enhance the accuracy and efficiency of SLR and SLT systems. Key techniques discussed include the use of deep learning frameworks such as Faster R-CNN, 3D-CNNs, and LSTMs, as well as hierarchical fusion models and skeleton-aware representations. Special attention is given to methods that address the challenges of precise action boundary detection, temporal cue learning, and robust key-point normalization. The paper also highlights the specific challenges encountered in different sign languages, such as the similarity of hand gestures in German sign language that differ only in lip shape. Through an analysis of these methods, the study seeks to offer a thorough grasp of the state-of-the-art in sign language technology.

Keyphrases: Hand Gesture, Sign Recognition, machine learning, sign language

@booklet{EasyChair:14820,
  author    = {Sheetal Patil and Amol Kadam},
  title     = {A Detailed Analysis of Recent Advances in Automatic Sign Language Recognition},
  howpublished = {EasyChair Preprint 14820},
  year      = {EasyChair, 2024}}