Human Activity Recognition with Discrete Cosine Transform in Lower Extremity Exoskeleton

Abstract

In this paper, in order to solve the problem of switching between different assisted state in exoskeleton, a human activity recognition scheme is proposed by using Discrete Cosine Transform(DCT) from accelerometer, gyroscope and plantar force data. In this method, plantar force data is used to determine the gait cycle and recognition window. After normalizing the acceleration and gyroscope data in the recognition window, DCT is used to find the features of the activity. Finally, the human-exoskeleton activity is classified by the Support Vector Machine(SVM) and different control strategies are used to different states. In experiments, the average recognition accuracy of the 7 human-exoskeletal systems commonly used for exercise (walking, up- stairs, down-stairs, standing, sitting, sit-to-stand, stand-to-sit) is 95.7% and identification window is 1.4s. Obviously, the introduction of plantar force sensors and accelerometers enables the recognition of human motion with higher accuracy and shorter recognition windows.