Unconstrained Lower Limb Motion Recognition by Angle Assisted Annotation-Based Surface EMG Sequential Labelling

Surface electromyography(sEMG)-based lower limb motion recognition has developed rapidly in a variety of applications and attracts extensive attentions. However, most of existing work focuses on single lower limb motion recogition(SLLMR) and ignores the importance of unconstrained lower limb motion recognition(ULLMR). In this paper, we propose a novel angle assisted annotation-based surface electromyography sequential labelling (A3-SESL) framework to solve ULLMR task. Specifically, A3-SESL is composed of labelling phase and learning phase. In term of labelling phase, due to difficulty of direct sEMG annotation, we introduce joint angle information to assist sEMG labelling. In term of learning phase, we first use labelled sEMG data to establish unconstrained lower limb motion sEMG (ULLM-sEMG) dataset, and then an light-weight temporal concolutional network is designed to perceive the latent difference information in supervised learning manner. Comprehensive comparision experimental results demonstrate the ability of proposed framework in recognizing lower limb motion.