Motion Pattern Recognition of Lower Limb Exoskeleton Based on SAPSO-SVM

Abstract

Accurate motion pattern recognition is the key to achieving human-computer cooperative control of the lower limb exoskeleton. This paper puts forward a motion pattern recognition method of lower limb exoskeleton based on an optimized support vector machine (SAPSO-SVM) via inertial sensor. This method introduces simulated annealing (SA) mechanism into particle swarm optimization (PSO) algorithm, which solves the problem that PSO algorithm is easy to converge locally to a certain extent, so as to acquire better parameters of classification model. Based on the RMS feature values of joint angle signals, we classified and recognized different lower limb motion patterns. The results reveal that the average recognition accuracy of SAPSO-SVM in single motion pattern is approximately 96.93%, and the Kappa coefficient is 0.9617, which has excellent consistency. The SAPSO-SVM method can further improve the effect of lower limb exoskeleton motion pattern recognition, and has good application value.