Abstract: To enhance the diversity and simplicity of hand gesture recognition, an approach based on ElectroMyoGraphy (EMG) and ACCeleration(ACC) signals was proposed to recognize dynamic gestures. Firstly, the gesture related information was collected by MYO sensors. Then, the dimensionality of ACC signal was reduced and the preprocessing of EMG was done. Finally, to reduce the number of training samples,the posture based on ACC signal was recognized by using Collaborative Sparse Representation (CSR) and the gesture based on EMG signal was classified by using Dynamic Time Warping (DTW) algorithm and the K-Nearest Neighbor (KNN) Classifier. When the ACC signal was identified by using CSR, the optimal number of samples and the dimensions of the dimensionality reduction were studied to reduce the complexity of gesture recognition. The experimental results show that the average recognition accuracy of the EMG for the hand gesture tested reaches 99.17%; the ACC signal for four postures achieve 96.88%. The recognition accuracy for the 12 dynamic gestures reaches 96.11%. This method has high recognition accuracy and fast calculation speed for dynamic gestures.

Key words: gesture recognition, Collaborative Sparse Representation (CSR), ElectroMyoGraphy (EMG), Dynamic Time Warping (DTW) algorithm, ACCeleration(ACC)

摘要： 为了增强手势识别的多样性和简便性，提出了一种基于肌电信号（EMG）和加速度（ACC）信息融合的方法来识别动态手势。首先，利用MYO传感器采集EMG和ACC的手势动作信息；然后分别对ACC和EMG信号作特征降维和预处理；最后，为减少训练样本数，提出用协作稀疏表示分类器来识别基于ACC信号的姿态手势，用动态时间规整（DTW）算法和K-最邻近分类器（KNN）来分类EMG信号的手形手势。其中在利用协作稀疏表示分类器识别ACC姿态信号时，通过对创建字典最佳样本个数以及特征降维的维数进行研究来降低手势识别的复杂度。实验结果表明，手形手势的平均识别率达到了99.17%，对于向上向下、向左向右4种姿态手势平均识别率达到 96.88%，而且计算速度快；对于总体的12个动态手势，其平均识别率达到96.11%。该方法对动态手势的识别率较高，计算速度快。

关键词: 手势识别, 协作稀疏表示, 肌电信号, 动态时间规整算法, 加速度