手势手语力效分析  

LMA Approach in Person-Independent Sign Language Recognition

作　　者：姜峰[1] 高文[1] 姚鸿勋[1] 陈熙霖[1] 

机构地区：[1]哈尔滨工业大学计算机科学与技术学院,哈尔滨150001

出　　处：《计算机学报》2007年第5期851-860,共10页Chinese Journal of Computers

基　　金：国家自然科学基金重点项目(60332010);新世纪优秀人才支持计划(NCET-05-0334)资助

摘　　要：在非特定人手语识别研究中,数据差异性带来的矛盾已使其成为一个亟待分析的问题.从人体运动学、语言学等角度对手语理解,是解决差异性矛盾进而推动非特定人手语识别的有效途径.文章以运动观测科学规则,特别是拉班的力效理论为基础,归纳了造成手语数据差异的因素,提出了手势手语力效要素的定义和描述方法;继而给出了非特定人手语数据的归整策略,规整后的数据用于训练与识别.在多种实验环境下进行的评估结果表明,识别的效果得到了明显的提升.In the research of singer independent sign language recognition, the contradiction brought about by the difference among data has made this difference an imperative problem to be analyzed. To understand sign language from the perspectives of human kinematics and linguistics is helpful to remove the individuality of various signers, at the same time ensuring the common characters of sign language words; this is an effective way to tackle the contradiction brought about by these differences. Based on the principles of movement observation science, especially the effort theory in LMA, taking into consideration the structurality of sign language, the paper describes and defines each dimension of effort in sign; then normalize sign language data of unspecific person to some extent, the normalized data is to be used in training and recognition. This method has been assessed under multiple experiment environments and been proved to greatly improve the recognition results.

关 键 词：手语识别 力效分析 非特定人 数据差异性 力效要素 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]