Classification of gait phases from lower limb EMG: application to exoskeleton orthosis

Abstract

This paper describes the use of Bayesian Information Criteria (BIC) along with some standard feature extraction methods and Linear Discriminant Analysis (LDA) classification algorithm to separate 8 different phases of gait by using electromyographic (EMG) signal data of the lower limb. Four time domain features along with 4th order Auto-Regressive model were used to get feature vector set from the EMG data of each leg of an able bodied person. Window of 50 ms (millisecond) was used such that it is within the controller delay limit. Then, the BIC segmentation algorithm was applied on the feature vector sets of 10 different gait cycles one by one to find out the locations of the boundaries between the phases. Due to the differences in the identified boundary locations for different gait cycles, the ambiguous part around each boundary was removed. The LDA classifier was then applied to the EMG feature vector set to classify 8 phases of gait. The classification accuracy increased by a significant amount in comparison to when BIC algorithm was not used. The work is our first step towards making an EMG signal driven foot-knee exoskeleton orthosis for the stroke patient having hemiparesis.