UACHR: Accurate CSI-Based Human Behavior Recognition Using Gaussian Goodness

Human motion and behavior analysis has become a new research field in pervasive computing, in which the feature modeling of behavior is particularly critical. This paper proposes an accurate CSI-based human behavior feature modeling method using Gaussian goodness. Firstly, this paper starts from the data distribution of CSI to find the difference between the still and action stages of the human body. By introducing Gaussian goodness of fit R-square, the degree of Gaussian fit of different stages is measured quantitatively, and the stage of action occurrence is extracted effectively. Secondly, in order to make full use of the CSI features of multiple antennas, a number of DTW-based FKNN classifiers are constructed to jointly judge behaviors at the level of neighboring samples. Experimental results show that the accuracy of the method is 95.33% and 91.33% respectively in the meeting room and the laboratory, and the system training time is greatly reduced compared with the KNN classifier.