基于单肌肉表面肌电-加速度融合的步态识别

Gait recognition method based on single muscle surface electromyography-acceleration fusion

为了解决步态识别中多传感器信息采集系统成本高、穿戴舒适性差、信息冗余以及系统复杂等问题,文章提出了一种简单的、低成本的基于单肌肉表面肌电-加速度融合的步态识别方法。首先搭建无线表面肌电-加速度信息采集系统,用于获取平地行走、上楼梯、下楼梯、上坡、下坡5种步态模式下的单肌肉表面肌电和加速度信号;选择股直肌作为试验相关肌肉获取信息,对采集到的信号预处理后提取时域、频域和时频域特征值,并采用核主成分分析(kernel principal component analysis,KPCA)方法对这些特征值信息融合;最后利用支持向量机(support vector machine,SVM)、反向传播(back propagation,BP)神经网络分别对融合后的特征向量进行分类,试验识别率分别为94.00%、91.33%。不同信息源步态识别准确率的分析结果验证了该文信息融合方法的有效性。

In order to solve the problems like high cost,wearing discomfort,information redundancy and system complexity of the multi-sensor information acquisition system in gait recognition,a simple and low-cost method for gait recognition based on single muscle surface electromyography-acceleration fusion is proposed.A system of surface electromyography and acceleration information acquisition in wireless is set up to obtain the single muscle surface electromyography and acceleration signals under five gait patterns of level-ground walking,stairs ascent,stairs descent,upslope and downslope.The rectus femoris is selected as the experimental relevant muscle acquisition information,and the acquired signals are preprocessed to extract the time domain,frequency domain and time-frequency domain eigenvalues.The kernel principal component analysis(KPCA)method is used to fuse these eigenvalue information and support vector machine(SVM)and back propagation(BP)neural networks to classify the fused feature vectors separately.The experimental recognition rates are 94.00%and 91.33%,respectively.By analyzing the accuracy of gait recognition from different information sources,the effectiveness of the information fusion is verified.