Gait Phase Discrimination during Kinematically Constrained Walking on Slackline

Abstract

We present a framework for the detection of swing and stance phases of gait cycle during walk on a perturbing platform, i.e., a slackline. The framework employs a time varying Kalman filter to reconstruct neural commands from the accelerometer data acquired from slackline perturbations that are generated by the participants. Electromyography (EMG) signals reconstructed from the estimated neural command were compared with surface EMG recordings during slackline walking. For four gait trials, the EMG signals were reconstructed with 92%-95% accuracy. These reconstructed signals were dimensionally reduced using principal component analysis (PCA) and classified using support vector machine (SVM), Decision Tree (DT), Naive Bayes (NB) classifiers. The stance and swing phases were discriminated, the performance of each classifiers was evaluated and classification accuracy was later enhanced with post processing. The DT algorithm outperformed others in detecting the stance and swing phases of gait cycle on slackline with 99.5% accuracy.