Accurate Gait Phase Detection using Surface Electromyographic Signals and Support Vector Machines

摘要

Modern components and materials allow creating increasingly complex, multi-functional prostheses with user-specific intelligent behavior. Complex behavior control, though, has to employ more accurate and precise models of the amputee and his prosthesis to be able to make use of the prosthesis' complete functionality. In this work, we concentrate on the accurate phase detection within a finegrained and state-full gait model for the continuous level gait. To this, we rely on four electromyography (EMG) sensors, placed at the thigh, and two force sensing resistors (FSR), placed below the heel and the toe. FSRs and a timed gait model are used to automatize EMG data recordings. Afterwards, gait model performance is verified using only EMG data. Here, we use support vector machines to detect muscular activity changes, indicating a new gait phase and therefore, a state switch within the gait model. We show that our approach generalizes well, even when using only 20 to 30 seconds for training. The gait model reaches accuracies of roughly 67% for an amputee and of 75% for a non-amputee individual when using a precise, seven phase level gait model.