Gait Phase Detection of Exoskeleton Robot Based on Optimized DAG-SVM

This paper proposes a gait phase detection method based on directed acyclic graph support vector machines (DAG-SVM) using weighted Euclidean distance optimization. Divide a gait cycle into six gait phases, including three stance phases and three swing phases. Heel, ball pressure, and knee and hip angle data fusion were used as input signals. When calculating the Euclidean distance between category samples, different coefficients are set for pressure data and angle data according to the category to which the gait phase to be classified belongs. The weighted Euclidean distance is obtained, and the topology of DAG-SVM is optimized according to the calculation results, so that it is applied to gait phase detection. This method can effectively solve the structural preference problem of DAG-SVM. Through experimental comparison, this method has higher detection accuracy than DAG-SVM with random structure.