Answers are grounded in this paper and cite their sources.
Video-based recognition of sign language is essential in improving communication and access for individuals who are deaf or have hearing impairments. The sign language encompasses two primary categories: single-hand and two-hand signs. Two-hand Thai sign language poses learning and communication challenges, including the difficulty of mastering hand coordination and the complexity in interpreting these gestures, especially by automated systems, leading to potential misunderstandings. To overcome this challenge, this paper presents a new novel large video dataset for two-hand poses of Thai Finger Spelling. Moreover, we propose the multimodal technique for RGB-sequencing-based modality on CNN-LSTM and VGG- LSTM models and the graph structure on skeleton modality using Temporal Graph Convolution Network (TGCN) under two scenarios: (1) static point-on-hand poses and (2) dynamic point-on-hand poses. Our result shows that the single modality is unsuitable for two-hand poses due to hand obstructions, which stem from insufficient coordinate sequence data or lack of feature extraction in the vision modality. To improve the accuracy and performance of two-hand postures, the study suggests combining RGB-sequencing with the vision modality and graph structure with the skeleton modality.
No takes yet. Share an insight, caveat, or question.
Vijitkunsawat et al. (2024) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: