Human Motion Intent Recognition Based on Kernel Principal Component Analysis and Relevance Vector Machine

For the low recognition rate of human motion intent, a human gait recognition method combining kernel principal component analysis (KPCA) and relevance vector machine (RVM) is proposed. The surface electromyography (sEMG) is selected as gait recognition information source, whose wavelet packet energy is extracted as characteristic value. The KPCA method is adopted to reduce the dimension of characteristic values for removing redundant information, so as to obtain the characteristic values which can reflect the human gait characteristics. Finally, the gait characteristic vectors are classified by RVM to recognize upslope, downslope, stairs ascent, stairs descent or level-ground walking. The feasibility and practicability of the method are verified through analyzing the gait recognition results of different subjects. Compared with BP (backpropagation) neural network and SVM (support vector machine) methods, the classification time of the proposed method is 2.6609×10^-4 s, and the recognition accuracy is 96.67%, which demonstrate it is an effective gait recognition method.