Lina Tong, Feng Zhang, Zeng-Guang Hou, Weiqun Wang, and Liang Peng
sEMG–joint angle estimation, rehabilitation, moving Butterworth filtering method, BP neural network, autoregressive (AR) model
Human motion estimation by surface electromyogram (sEMG) is one of the most important human intention recognition methods for active rehabilitation training. This paper proposes a back propagation (BP) neural network and autoregressive (AR) model based real-time sEMG–joint angle estimation method. To reduce the time delay, a moving Butterworth filtering method is designed to filter the lower limb multi-channel sEMG signals. Then correlation analysis between sEMG signals and joint angles is made to reduce redundant channels. A first-order BP neural network is used to build the mapping relationship between multi-channel sEMG signals and joint angles, then the approximated angle by BP model is adjusted by the AR de-noising model, which describes the angle variation features of the given training mode to improve the accuracy and continuity. To validate this method, five able-bodied subjects participated in cycling exercise experiment, and the angle estimation results show that this method presents a good performance on real-time computation, accuracy and continuity.
Important Links:
Go Back
