Moving Target Localization and Activity/Gesture Recognition for Indoor Radio Frequency Sensing Applications

In this paper, a dual-frequency continuous wave radar is proposed to achieve both localization and activity/ gesture recognition simultaneously. Specifically, features of different movements will be classified by the activity and gesture recognition network (AGRNet) which is a lightweight network based on MobileNet. The data that are recognized corresponding to walking will be used for moving target localization by comparing the phase difference in the Doppler domain between dual frequencies. In addition, a segmentation method is proposed to effectively segment continuous signals into individual time-periods corresponding to different motions by detecting the boundaries of signal changing. The experimental results show that the proposed method accomplishes the classification accuracy over 91% with 8 motion classes with a localization accuracy in the centimeter level.