本文研究了基于肌电(Electromyography,EMG)-惯性融合的人体运动估计问题,提出了一种序贯渐进高斯滤波网络(Sequential progressive Gaussian filtering network,SPGF-net)估计方法来形成肌电和惯性的互补性优势,以提高人体运动估计精度和稳定性.首先,利用卷积神经网络对观测数据进行特征提取,以及利用长短期记忆(Long short-term memory,LSTM)网络模型来学习噪声统计特性和量测模型.其次,采用序贯融合的方式融合异构传感器量测特征,以建立高斯滤波与深度学习相结合的网络模型来实现人体运动估计.特别地,引入渐进量测更新对网络量测特征的不确定性进行补偿.最后,通过实验结果表明,相比于现有的卡尔曼滤波网络,该融合方法在上肢关节角度估计中的均方根误差(Root mean square error,RMSE)下降了13.8％,相关系数(R2)提高了4.36％.
Human Motion Estimation Based on EMG-Inertial Fusion:A Gaussian Filtering Network Approach
This paper investigates the issue of human motion estimation based on the fusion of electromyography(EMG)and inertial data.A sequential progressive Gaussian filtering network(SPGF-net)is proposed to leverage the complementary advantages of EMG and inertial data for enhancing the accuracy and stability of human mo-tion estimation.First,a convolutional neural network is employed to extract features from the observed data and a long short-term memory(LSTM)network model is utilized to learn the statistical properties of noise and the meas-urement model.Second,a sequential fusion method is adopted to fuse the measurement features from heterogen-eous sensors,thus a combined network model that integrates Gaussian filtering with deep learning techniques is formed for human motion estimation.Moreover,a progressive measurement update is introduced to compensate for the uncertainty in the network's measurement features.Finally,experimental results indicate that,compared with existing Kalman networks,the proposed fusion method has a 13.8％ reduction in root mean square error(RMSE)and a 4.36％ increase in the coefficient of determination(R2)for upper limb joint angle estimation.
万方数据
