Body Gestures Recognition Based on CNN-ELM Using Wi-Fi Long Preamble

Recently, researchers around the world have been striving to develop human–computer interaction systems. Especially, neither special devices nor vision-based activity monitoring in home environment has become increasingly important,and has had the potential to support a broad array of applications. This paper presents a novel human dynamic gesture recognition system using Wi-Fi signals. Our system leverages wireless signals to enable activity identification at home. In this paper, we present a novel Wi-Fi-based body gestures recognition model by leveraging the fluctuation trends in the channel of Wi-Fi signals caused by human motions. We extract these effects by analyzing the long training symbols in communication system. USRP-N210s are leveraged to set up our test platform, and 802.11a protocol is adopted to implement body gestures recognition system. Besides, we design a novel and agile segmentation algorithm to reveal the specific pattern and detect the duration of the body motions. Considering the superiority of feature extraction, convolutional neutral networks (CNN) is adopted to extract gesture features, and extreme learning machine (ELM) is selected as classifier. This system is implemented and tested in ordinary home scenario. The result shows that our system can differentiate gestures with high accuracy.