PSO-SVM-based gait phase classification during human walking on unstructured terrains: Application in lower-limb exoskeleton

Gait analysis is of great importance to ensure that gait phases induced by robotic exoskeleton are tailored to each individual and external complex environments. The objective of this work is to develop a pressure insole system with redundant functionality for gait phase classification based on the analysis of ground reaction force on unstructured terrains. A support vector machine optimized by particle swarm optimization was proposed for classifying four gait phases including initial contact, mid stance, terminal stance and swing phase. Seven pressure sensors are employed according to the plantar distribution contour of ground reaction force and walking data acquisition is conducted on treadmill, concrete pavement and wild grassland, respectively. Two classifiers, support vector machine-based classifier I and PSO-SVM-based classifier II are constructed on the basis of gait data set obtained on treadmill. Experimental results showed that the proposed PSO-SVM algorithm exhibits distinctive advantages on gait phase classification and improves the classification accuracy up to 32.9%–42.8%, compared with that of classifier based solely on support vector machine. In addition, some unwanted errors, intentional attacks or failures can be successfully solved with fast convergence rate and good robustness by using particle swarm optimization.