Abstract: A gesture recognition approach based on Dynamic Time Warping（DTW） algorithm is proposed in an attempt to overcome some critical problems with the currently available hand gesture recognition methods using electromyography（EMG）, including low accuracy, lack of real-time. Firstly, the raw EMG data is segmented by the overlapped sliding averaged energy to detect active actions. Then, the feature data of segmentation is extracted by using Mean Absolute Value（MAV）. Finally, the EMG signal of eight dimensions is fused and the gesture recognition is obtained by matching the test samples with the template in DTW method. The template making is devised by finding the warping paths, in order to realize the inter-individual hand gesture recognition. The results show that the recognition accuracy of the EMG signals for the hand gestures tested can reach as high as 96.09%, and the DTW based approach is fast and is potentially a promising choice for real-time hand gesture recognition applications.

Key words: human-computer interaction, Myo sensor, electromyography（EMG） signal, Dynamic Time Warping（DTW） algorithm, gesture recognition, template making

摘要： 为了提高肌电信号手势识别算法的准确度，增强实时性，提出了一种基于动态时间规整（DTW）算法的手势识别方法，该方法利用肌电信号（EMG）对个体间的手势进行识别。首先，采用滑动平均能量的方法对原始的EMG信号进行数据分割，探测有效动作；其次，对于分割的数据段使用平均绝对值（MAV）来提取信号特征；最后，用DTW算法将8维的EMG信号融合并计算测试样本和模版的相似度，其中采用了DTW算法寻找规整路径的方法进行了模板制作，实现了个体间的手势识别。实验结果表明，使用DTW算法对肌电信号进行手势识别，其动作识别的准确率达到96.09%，该方法计算速度快，实时性强。

关键词: 人机交互, Myo传感器, EMG信号, 动态时间规整（DTW）算法, 手势识别, 模板制作