A real-time leg motion recognition system by using Mahalanobis distance and LS_SVM

With the increasing requirements of the society to help those with special needs (e.g., physically disabilities, the old and the injured individuals), lower limb rehabilitative robot has been expected to have a significant potential foreground. Surface electromyography (sEMG) signal will be utilized as the intention command to control the lower limb assisting robot in this research. Six types of leg movements, collected by placing electrodes on four appointed muscles, are involved. In order to realize on-line controlling, the recognition accuracy and the amount of data are two critical factors. Comparing various feature extraction approaches in time domain and time-frequent domain, this paper proposes a real-time control system with 99.44% identification rate and low dimension feature vectors that are selected by Mahalanobis distance (MD). Furthermore, a specific least squares support vector machine (LS_SVM) is designed to conduct the classification task in this context.