Human Activity and Gesture Recognition based on WiFi

Due to the wide availability of WiFi devices in almost every modern building, activity and gesture recognition systems using WiFi have recently been in the spotlight. Given the low cost and touch-free working of WiFi, this sensing approach has found success in various fields, including elder care, child safety, and smart homes that enable householders to monitor everything. This paper aims to investigate the human activity and gesture recognition scheme that uses the Channel State Information (CSI) value provided by WiFi devices. To achieve high accuracy in the measurement, firstly, a median filter is used to filter out the noise from Channel State Information extracted from the WiFi signal. Secondly, massive features are extracted to present intrinsic characteristics for each gesture and activity. Finally, Random Forest (RF) and Support Vector Machine (SVM) algorithms with cross-validation techniques are used for data classification. The system is implemented on a public dataset (WiAR) involving ten volunteers, each of whom performs 16 activities in each experiment. Experimental results show an average recognition accuracy of 92% and 91% for Random Forest and Support Vector Machine, respectively, which outperforms the existing system by 3.14 percent for RF and 1.14 percent for SVM.