Physical Activity Recognition Using Continuous Wave Radar With Deep Neural Network

Abstract

In this study, we investigated the feasibility of using a continuous-wave radar sensor for detecting physical activities. The transfer learning method, applying a pre-trained deep neural network (Alexnet), was used to perform the classification task. Doppler signatures of these activities were converted to spectrogram figures as the input of the classifier. The classifier was tested in five-fold cross-validation and leave-one-person-out. The Fl-score of five-fold cross-validation had higher score, which ranged from 71.11 % to 82.05%.