Discrimination of waist motions based on surface EMG for waist power assist suit using support vector machine

This paper presents a signal processing for discrimination of waist motions including forward and backward bendings and right and left twists. The system is planed to implement to a waist power assist suit that physically helps a caregiver in personal care tasks. The motion discrimination is based on surface electromyogram (SEMG) of right and left erector spinae muscles that dominate the motions of interest, and accomplished by using four SVMs in which each SVM is a binary classifier for each of four motions. We construct a strong multi-class classifier based on combination use of four SVMs. With a peripheral FFT-based prefilter, the start point of motion is estimated, and employed for a trigger to calculate a feature vector. We show that the proposed processing has a promising discrimination and false-positive rates for implementation. In addition, we summarize some essential problems to improve the performance of the system as future works.