Periodic Locomotion-model Recognition Based on Electromyography of Thigh Stump

Abstract

In view of the problem of continuous movement pattern recognition for above-knee prostheses control, a periodic locomotion-model recognition method was proposed based on electromyography of thigh stump. Firstly, after analyzing the surface electromyography of gluteus maximus, multi-feature sections detection algorithm was proposed based on moving windows, and multi-feature sections within a motion cycle were extracted. Secondly, random forest algorithm was applied to recognize the movement pattern of each section. Finally, a periodic pattern recognition method based on binary tree was proposed to fuse the recognition results of each section. The experiment results indicated that this method improved the recognition accuracy by about 8% with multi-feature sections fusion. The pattern recognition of periodic motion (flat walking, upstairs, and downstairs) and aperiodic motion (sitting and standing) were realized, and the recognition accuracy and real-time performance have improved obviously.