Human Counting and Action Recognition with WiFi via Deep Learning

For simultaneously recognizing the indoor people's number and their actions, a system based on deep learning for human counting and action recognition using channel state information (CSI) of WiFi signal is proposed. A phase difference correction algorithm was proposed for eliminating the skipping of the original phase difference and a deep residual network was used to extract the representative features of the classification. The experimental results show that the accuracy of this method in the recognition of people's number is 99%, and the average accuracy in the recognition of two-person actions is 95.5%, which has a good recognition effect.