A Novel Anomaly Detection Algorithm Based on OCSVM for Lower Limb Exoskeleton System

This paper presents a novel anomaly detection algorithm for lower limb exoskeleton system, which based on the gait characteristics of human walking on a Hat ground. This algorithm is suitable for lower limb exoskeleton system which can obtain the angles of hip, knee and ankle in real time. Previous studies have shown that the traditional balance evaluation methods such as zero moment point (ZMP), center of press (COP) and base of support (BOS) are difficult to solve and rely on accurate plantar pressure sensors. Therefore, through the analysis of the characteristics of human body stability, this paper uses the position relationship between the center line of feet (CLOF) and center of mass (COM) as the evaluation index of human body balance. To verify the correctness of this index, the forward, backward, left and right lateral abnormal data of human body are collected by inertial navigation, and the algorithm is used to train the anomaly classifier. Finally, through the test of normal sample and abnormal sample data, the classification accuracy is 99.821% and 99.297% respectively. In the next step, the algorithm will be used to control the balance recovery of the lower limb exoskeleton system.