Scholarlink

<p>Colombian Sign Language (LSC) is part of Colombia's cultural heritage and is protected by the state as the official language of people with hearing disabilities. Despite the tools implemented by the government, significant social inclusion gaps persist, leading to discrimination due to a lack of language knowledge and the absence of interpreters. This article presents a system for dynamic alphanumeric interpretation of LSC, using spelling-based recognition for both words and numerical quantities. The methodology is based on CRISP-DM, and the LSC70 database was used for alphanumeric signs in two types of images: full-body and hand-focused. Two modeling approaches were followed using deep learning. The first approach is image-based, where raw image inputs are processed through CNNs combined with a BILSTM network to capture spatial and temporal features. In contrast, the second approach is coordinate-based, where hand landmarks are extracted from images using MediaPipe, focusing instead on the temporal patterns of landmark movements, which are processed with a BILSTM network. The results showed that the combined models achieved lower accuracy, with no convergence for the full-body approach but reaching 75.9% accuracy for the hand-focused approach. However, the coordinate-based models outperformed these results, achieving 75.5% and 87.6% accuracy for the full-body and hand-focused approaches, respectively. The latter model was deployed on a web platform created using Streamlit for real-time interpretation. In conclusion, a dynamic online alphanumeric LSC interpreter was developed to promote communication and social inclusion for people with hearing disabilities by enabling effective communication and integration.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Dynamic interpretation system for alphanumeric signs of the Colombian Sign Language -LSC- through the combination of deep learning networks

  • 期刊论文

  • 发表日期：

    2025年6月18日

查看全文
  • Jader Alejandro Muñoz Galindez,
  • Rubiel Vargas Cañas

摘要

Colombian Sign Language (LSC) is part of Colombia's cultural heritage and is protected by the state as the official language of people with hearing disabilities. Despite the tools implemented by the government, significant social inclusion gaps persist, leading to discrimination due to a lack of language knowledge and the absence of interpreters. This article presents a system for dynamic alphanumeric interpretation of LSC, using spelling-based recognition for both words and numerical quantities. The methodology is based on CRISP-DM, and the LSC70 database was used for alphanumeric signs in two types of images: full-body and hand-focused. Two modeling approaches were followed using deep learning. The first approach is image-based, where raw image inputs are processed through CNNs combined with a BILSTM network to capture spatial and temporal features. In contrast, the second approach is coordinate-based, where hand landmarks are extracted from images using MediaPipe, focusing instead on the temporal patterns of landmark movements, which are processed with a BILSTM network. The results showed that the combined models achieved lower accuracy, with no convergence for the full-body approach but reaching 75.9% accuracy for the hand-focused approach. However, the coordinate-based models outperformed these results, achieving 75.5% and 87.6% accuracy for the full-body and hand-focused approaches, respectively. The latter model was deployed on a web platform created using Streamlit for real-time interpretation. In conclusion, a dynamic online alphanumeric LSC interpreter was developed to promote communication and social inclusion for people with hearing disabilities by enabling effective communication and integration.