基于多传感器信息检测和融合的中国手语识别研究[].doc

29卷 5期 中国生物医学工程学报 Vol. 29 No. 5 2010年 10月Chinese Journal of Biomedical Engineering October 2010 基于多传感器信息检测和融合的中国手语识别研究 王文会陈香*阳平李云杨基海 (中国科学技术大学电子科学与技术系，合肥 230027) 摘要:有效融合加速计、摄像头和表面肌电 3种低成本传感器在手势动作信息捕获上的优势，是提高手语手势识 别率和种类的重要研究内容。提出一种基于多传感器信息检测和融合的中国手语识别方法 :先利用表面肌电的幅 值信息，对 3类传感器信号进行手势分割，并实现单双手词的划分 ;然后借助视觉信号，完成有遮挡和无遮挡双手 词的划分 ;最后利用 Sugeno模糊积分，实现不同特征匹配结果的决策融合。结果表明，对 4位受试者、201个高频 手语词开展手势识别实验，其识别率均在 99%以上，证明该基于多传感器信息检测和融合的手势识别方法在中国 手语识别上的有效性。 关键词 :手语识别 ;多传感器融合 ;隐马尔科夫 ;模糊积分 中图分类号 TP391. 4文献标识码 A文章编号 0258-8021(2010)05-0665-07 Chinese Sign Language Recognition Based on Multiple Sensors Information Detection and Fusion WANG Wen-Hui CHEN Xiang* YANG Ping LI Yun YANG Ji-Hai (Department of Electronic Science and Technology，University of Science＆ Technology of China，Hefei 230027，China) Abstract: The efficient fusion of hand gesture information captured by a three-axis accelerometer，a webcam and four surface electromyography sensors is an important research field for improving the performance of sign language recognition system. In this paper，a multi-sensor information detection and fusion method was proposed. Firstly，the amplitude information of myoelectric signal was utilized to extract active segments of hand gestures and divide sign gestures into single-hand type and double-hand type. Then double-hand sign words were further classified into occlusion or non-occlusion class by vision signal. Lastly，decision-level fusion approach with Sugeno fuzzy integral was applied on local matching results of multiple classifiers for improving classification performance. Experimental results for 201 high-frequency sign words from 4 signers obtained the classification accuracy of more than 99%，indicating the effectiveness of the proposed method for Chinese sign language recognition. Key words:sign