摘要
手语是聋人使用的语言 ,是由手形动作辅之以表情姿势由符号构成的比较稳定的表达系统 ,是一种靠动作 /视觉交际的语言 .手语识别的研究目标是让机器“看懂”聋人的语言 .手语识别和手语合成相结合 ,构成一个“人-机手语翻译系统”,便于聋人与周围环境的交流 .手语识别问题是动态手势信号即手语信号的识别问题 .考虑系统的实时性及识别效率 ,系统选取 Cyberglove型号数据手套作为手语输入设备 ,并采用了 DGMM(dynamic Gaussianm ixture m odel)作为系统的识别技术 ,即利用一个随时间变化的具有 M个分量的混合 Gaussian N-元混合密度来模型化手语信号 ,可识别中国手语字典中的 2 74个词条 ,识别率为 98.2 % .与基于 HMM的识别系统比较 ,这种模型的识别精度与 HMM模型的识别精度相当 ,其训练和识别速度比 HMM的训练与识别速度有明显的改善 .
Sign language is the language used by the deaf, which is a comparatively steadier expressive system composed of signs corresponding to postures and motions assisted by facial expression. It is communication using motion/vision. The objective of sign language recognition research is to “see” the language of the deaf. The integration of sign language recognition and sign language synthesis jointly comprise a “human computer sign language interpreter”, which facilitates the interaction between the deaf and their surroundings. The issue of sign language recognition is to recognize dynamic gesture signal, that is, to recognize sign language signal. Considering real time property and recognition performance of the system, Cyberglove is selected as the gesture input device in the system under discussion and DGMM(dynamic Gaussian mixture model) is used as a recognition technique, which models sign language signal by a time varying density function composed of M N Gaussian mixture density function. The system can recognize 274 sign language words coming from the dictionary of Chinese sign language with the accuracy of 98.2%. Compared with the recognition system based on HMM, the recognition rate of DGMM is nearly equal to that of HMM, and the training and recognition speed of DGMM is apparently much faster than that of HMM.
出处
《计算机研究与发展》
EI
CSCD
北大核心
2000年第5期556-558,共3页
Journal of Computer Research and Development
基金
国家"八六三"高技术研究发展计划基金项目!(项目编号 863 -3 0 6-ZT0 3 -0 1-2 )
国家自然科学基金项目!(项目编号 697893 0 1)
国
关键词
隐式马尔可夫模型
手语识别系统
DGMM
sign language recognition, dynamic Gaussian mixture model, hidden Markov model