Using UDP to Realize Flexible and Portable Human Activity Recognition

With the development of smart home, the research of human activity recognition (HAR) in home scene has been attracting more and more attentions in both the academic and industrial fields. Benefited from the widespread deployment of Wi-Fi routers in every household, research on Wi-Fi-based human activity recognition is cost-effective and feasible. Human activity recognition can be realized by obtaining the changes in the Wi-Fi channel state caused by human activities in the environment, and then using machine learning models for corresponding training. In this paper, a low-cost HAR system based on ESP32 is proposed, which can collect and transmit channel state information (CSI) data flexibly and efficiently by using a user datagram protocol (UDP) communication transceiver server. The proposed system avoids the use of serial port communication or SD card to collect CSI data, which reduces the difficulty of CSI data collection. An actual verification of the method is built and the performances of several common training models are compared. Our initial results show that the convolutional neural network (CNN) provide the best performances, i.e., reaching an accuracy rate of 98.6%.