Human Motion Recognition Based On SVM Manifold Regularization

With the rapid development of artificial intelligence,human action recognition technology has attracted much attention.Traditional motion recognition technologies,based on wearable sensor devices and video-based methods,are limited by the high cost and poor privacy.Recently,the action recognition technology based on wireless signal has attracted the attention of researchers.This method has the advantages of high popularity,low cost,strong concealment,strong environmental adaptability and so on.It can achieve better recognition results by the collaboration with the traditional method.Human motion can be recognized by the analysis of the channel state information(CSI)of wireless signals.The main contribution of this thesis can be summarized as follows:A new idea is proposed in this paper which extract features from the data set of human motion by the method of vision.Our analysis shows that after converting denoising CSI data into images,the texture of the same kind of action images is similar,meanwhile there is a great difference in different kinds of action images.Therefore,Gabor and scale invariant feature transformation are used to extract relevant features that can reflect image texture.The collected CSI data is added to the support vector machine model by the traditional time domain method and the feature values extracted by the visual method to perform classification training,which can effectively improve the motion recognition accuracy.In view of the fact that unlabeled data is easily accepted in CSI-based action recognition,this paper proposes a semi-supervised SVM model based on manifold and Hessian regularization.Similar labels are assigned to close samples by the method of manifold.Hessian can convert part of the unlabeled data intolabeled data and train the model with the original labeled data.In this paper,these two rules are introduced into the SVM model in the form of regularization terms to form a new semi-supervised model.In this paper,the improved classification model is compared with the traditional support vector machine model for classification experiments.The experimental results show that the improved classification model can significantly improve the performance of motion recognition.