加速计(ACC)和表面肌电(SEMG)传感器是两种有效轻便的手势捕获设备.本研究提出一种采用多级决策树融合ACC和SEMG信息识别手语词根,并引入统计语言模型进行词根接续判断和错误纠正的中国手语连续语句识别方法.对包含有120个词根的200组连续中国手语句子展开的识别实验结果表明,该方法可以有效的从连续信号中识别出词根序列,120个手语词根全局平均识别率接近95%,句子识别率接近90%,采用纠错模型的方法与未采用纠错模型相比,词根的平均识别率提高了4%左右,句子识别率提高了10%.这种结合模式识别和自然语言处理的手语识别方法在连续手语识别和人机交互领域有着广阔的应用前景.%Accelerometer (ACC) and surface electromyography (SEMG) sensors are two effective portable devices to capture gestures. In this paper, a multi-sensor information fusion method was proposed to recognize the Chinese sign language gestures. Firstly, a hierarchical decision tree was constructed for the information fusion of ACC and EMG signals to recognize the subwords of Chinese Sign Language (CSL). Then the statistical language model was constructed to detect and correct error in the process of the recognition. For the recognition of 120 CSL subwords and 200 sentences, the average recognition accuracies of our method could up to 91％ and 84％ respectively. The comparative analysis of experimental results showed that the statistical language model could improve the recognition accuracies of 120 CSL subwords by 9％ and the recognition accuracies of 200 sentences by 13％. The results indicated that the proposed method could effectively recognize Chinese sign language.
展开▼
