【摘要】 目前手语识别研究的难点之一在于如何实现大词汇量的连续语句识别 .针对这个问题 ,该文提出了几个有效的方法 ,包括修正转移矩阵、状态结点的捆绑、快速匹配、在搜索路径中加入词跳转的估计参数等 .利用上述技术 ,该文实现了一个基于数据手套和位置跟踪器的大词汇量的连续的中国手语实时识别系统 .对中国手语辞典中收录的 5 10 0个词以及一批连续语句作实验 ,实验结果表明 ,文中所介绍的技术在提高系统识别速度和准确率方面都很有效更多还原

【Abstract】 Gestures are destined to play an increasingly important role in human computer interaction in the future. Closely related to the field of gesture recognition is the field of sign language recognition. Sign languages are the primary mode of communication for many deaf people, and because they are full fledged languages in their own rights, they offer a much more structured and constrained research environment than general gestures. Furthermore, working on sign language recognition systems would also make the interaction of deaf people with their surroundings easier. Most previous work has focused on isolated sign language recognition and continuous recognition based on small set of signs. Hitherto, one of the major challenge that faces Sign Language recognition is to develop methods that will scale well with increasing vocabulary size. In this paper, a novel approach to continuous, whole sentence Chinese Sign Language (CSL) recognition based on large vocabulary is presented. In order to improve the recognition speed and accuracy, some useful new ideas are proposed in this paper, including modifying the transferring probability, states tying, fast matching and using the estimating parameter of transferring between signs in search algorithm. The aim of state tying is to reduce computation load in decoding process. Fast matching is used to decide which signs is the potential one. Adding a parameter can weaken the effect of the extra movement between two signs. Using these techniques, a real time system designed for recognizing continuous Chinese Sign Language sentences with a large vocabulary based on two CyberGloves and a tracker is built. Experiments are carried out on 5100 signs included in Chinese Sign Language Dictionary. 1000 continuous sentences are test. These sentences are made randomly based on the vocabulary. The results show that the techniques proposed in this paper are efficient on either recognition speed or recognition rate. The sentences can be recognized in real time.更多还原