A deep Kalman filter network for hand kinematics estimation using sEMG

In human-machine interfaces (HMI), deep learning (DL) techniques such as convolutional neural networks (CNN), long-short term memory (LSTM) and the hybrid CNN-LSTM framework have been exploited for hand kinematics estimation using surface electromyography (sEMG). However, these DL only capture relationship between sEMG kinematics, but ignores prior knowledge of system. By contrast, Kalman filter (KF) can apply gain to combine internal transition model observation effectively. To this end, we propose a novel architecture named network (DKFN), in which utilize CNN extract high-level features from employ LSTM-based process (LSTM-KF) conduct sequential regression. particular, LSTM-KF adopts computational graph KF estimates parameters transition/observation data LSTM modules. With process, advantages be jointly. Experimental results demonstrate that proposed DKFN outperform regression wrist/fingers estimation.