非特定人手语数据生成及其有效性检测  

Data Generation and Its Validity Inspection of Signer-Independent Sign Language

作　　者：倪训博[1] 赵德斌[1] 高文[1,2] 姜峰[1] 姚鸿勋[1] 

机构地区：[1]哈尔滨工业大学计算机科学与技术学院,黑龙江哈尔滨150001 [2]中国科学院计算技术研究所,北京100190

出　　处：《软件学报》2010年第5期1153-1170,共18页Journal of Software

基　　金：国家自然科学基金(Nos.60603023;60533030);北京市自然科学基金No.4061001~~

摘　　要：根据手势手语的特点,提出了手语语言学和人体运动学相结合的非特定人手语数据的生成和检测方法.首先,Mean-Shift算法有控制生成强度的优点,将改进的Mean-Shift算法应用于手形数据通道的生成,以保持手势手语的语言学特性,并应用关键手形的音韵标记进行数据有效性的检测;其次,为了丰富手语手势动作的运动特性,将改进的遗传算法应用于与运动相关的数据通道进行数据生成,并应用拉班舞谱对其进行数据有效性检测;最后,提出了基于原始样本的检测实验框架,使得所提出的检测方法适用于语言类的多类别数据检测问题.实验结果表明,所提出的非特定人手语数据的生成和检测方法是有效的.This paper proposes the combination of sign language linguistics with human kinematics to generate and detect the data of SISL （signer independent sign language） according to the characteristics of gesture sign language （GSL）. An improved Mean-Shift algorithm is applied to the generation of hand shape data channels without losing the linguistic features of GSL, and then the key hand shape phonetic notation is used to detect the effectiveness of data. In order to enrich the kinematic characteristics of GSL, an improved genetic algorithm is applied to the generation of movement related data channels. Moreover, Labannotation is adopted to inspect the effectiveness of data. Finally, an experimental inspection framework is established based on an original sample to make the proposed detection method adapt to multi-classes data inspection of linguistics. Experimental results show that the proposed method for the generation and detection of SISL data is effective and feasible.

关 键 词：手语识别 音韵标记 拉班舞谱 手语语言学 人体运动学 MEAN-SHIFT算法 遗传算法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]