中国手语识别方法及技术综述  

Review of Chinese Sign Language Recognition Methods and Technologies

作　　者：蒋贤维 孙计领 张艳琼 王立平 蒋小艳 韩雪 IANG Xianwei;SUN Jiling;ZHANG Yanqiong;WANG Liping;JIANG Xiaoyan;HAN Xue(School of Mathematics and Information Science,Nanjing Normal University of Special Education,Nanjing,210038)

机构地区：[1]南京特殊教育师范学院数学与信息科学学院,南京210038

出　　处：《现代特殊教育》2024年第6期47-58,共12页Modern Special Education

基　　金：国家社会科学基金年度项目“残疾人融合发展评价及提升路径研究”(20BRK029)研究成果

摘　　要：中国手语具有自己独特的文化内涵和复杂表达,是近3000万听障人士融入社会的重要手段。手语识别技术能帮助听障人士走出信息孤岛,和健听人建立有效沟通。中国手语识别方法大致经历了传统技术识别和现代智能识别两个时期。前者主要包含数据收集、预处理、特征提取和分类识别四个主要阶段,主流技术有HMMs、SVM和DTW等,基于手语手形数据完成识别,不依赖海量样本数据;后者主要利用深度神经网络和人工智能技术,强调深度学习,迁移学习和技术融合,模型对样本数据量的依赖程度较高。我国已经开始广泛建设各类手语语料库,但需要进一步规范和推广。Chinese Sign Language has its own unique cultural connotations and complex expressions,and it is an important means for more than 30 million hearing-impaired people to integrate into society.Sign language recognition technology can assist individuals with hearing impairments in bridging communication gaps and establishing effective communication with those who can hear.Chinese sign language recognition methods have gone through roughly two stages of traditional technology recognition and modern intelligent recognition.The former mainly includes four stages of data collection,preprocessing,feature extraction,and classification recognition.Hidden Markov Models(HMMs),Support Vector Machines(SVM),and Dynamic Time Warping(DTW)are mainstream technologies.It achieves recognition based on hand data without relying on extensive sample data.The latter mainly combines deep neural network and artificial intelligence technology,emphasizing deep learning,transfer learning,and technology integration.The model is highly dependent on the amount of sample data.China has started to extensively develop various sign language corpora,but it requires further standardization and promotion.

关 键 词：手语识别技术 语料库 深度神经网络 迁移学习 

分 类 号：G760[文化科学—特殊教育学]