机器学习作为手语识别系统开发工具：综述

文献信息

中文摘要

听力损失是一项全球性挑战，影响聋人群体沟通、学习和工作的能力。目前，世界卫生组织报告称超过4.3亿人患有致残性听力损失。在康复替代方案中，手语是应用最广泛的方式之一；然而，聋人与听人群体之间的有效沟通仍然存在障碍。为帮助弥合这一沟通鸿沟，技术工具的开发得到推动，尤其是手语识别系统（SRS）。本综述的目的是对当前SRS研究现状进行整合分析，这些研究由手语教学与学习应用所驱动，识别从数据采集和特征提取到分类和技术部署的整个流程中的主导方法学方法，并强调高实验性能与现实世界可及性之间持续存在的差距。本文献综述分析了55篇文章，这些文章选自2020年至2025年间Scopus数据库索引期刊中初步识别的161条记录。分析显示，这些系统通常分为两个阶段：特征提取和分类。最后，尽管SRS取得了高性能，但在将这些系统转化为最终用户可有效使用的技术方面仍存在差距。

英文摘要

Hearing loss represents a global challenge, affecting the ability of the deaf community to communicate, study, and work. Currently, the World Health Organization reports that more than 430 million people live with disabling hearing loss. Among rehabilitation alternatives, sign language is one of the most widely adopted; however, effective communication between deaf individuals and the hearing population remains a barrier. To help bridge this communication gap, the development of technological tools has been promoted, particularly Sign Recognition Systems (SRSs). The purpose of this review is to provide an integrative analysis of the current state of SRS research motivated by sign language teaching and learning applications, identifying dominant methodological approaches across the full pipeline-from data capture and feature extraction to classification and technological deployment-and highlighting the persistent gap between high experimental performance and real-world accessibility. This literature review analyzes 55 articles, selected from an initial pool of 161 records identified in journals indexed in the Scopus database between 2020 and 2025. The analysis revealed that these systems are generally structured in two stages: feature extraction and classification. Finally, despite the high performance achieved by SRS, a gap remains in transferring these systems into technologies that can be effectively used by end users.