Joint Recognition Of Human Target Location And Activity Based On Multidimensional Features Of CSI Images

With the widespread deployment of wireless networks and the popularity of wireless signals,the joint location and behavior recognition technology based on CSI(Channel State Information)has a wide application prospect in the fields of human-computer interaction,intelligent home,remote health monitoring,security alarm and so on.The traditional human target location and behavior recognition technologies are based on Computer Vision Technology,wearable sensor technology and radio frequency identification technology,etc.,many sensors and cameras are needed to collect data,which is not only expensive and difficult to deploy,but also interfered by environmental lighting and obstructions.In contrast,CSI-based human target location and behavior identification technology has the advantages of no additional deployment of equipment,convenient data collection,insensitive to environmental lighting and blocking factors,and so on,this technology has already become the core technology which has the development prospect very much in the intelligent perception domain.Currently,CSI-based human target single-dimensional information recognition technology has achieved high recognition accuracy,but multi-dimensional information recognition technology only uses the amplitude information of a communication link subcarrier for complex multi-dimensional information recognition,without fully utilizing the fine-grained temporal and spatial information of CSI.Therefore,new algorithms are needed to further improve recognition accuracy.In this thesis,a Joint Recognition of Location and Action algorithm based on CSI image multi-dimensional features(JRLA-MFCI)is proposed.The algorithm fully utilizes the fine-grained information of CSI to construct an antiinterference CSI amplitude difference and phase difference image,and then extracts texture and color features from the aforementioned CSI images to form a feature vector.Finally,a support vector machine(SVM)algorithm with a quadratic polynomial kernel function and optimized parameters is used for joint recognition of human target position and behavior.A large number of comparative experiments and analyses show that the CSI images constructed in this thesis,the extracted multi-dimensional features,and the SVM algorithm’s kernel function ensure that JRLA-MFCI algorithm has high accuracy and robustness in joint recognition of position and behavior.