Human behavior detection and recognition method based on Wi-Fi signals

Abstract

With the development of WLAN technology, human motion and behavior analysis has become an emerging research field in ubiquitous computing. Existing works usually only use time-related features or space-related features, and lack effective means to integrate and utilize both at the same time, which makes it difficult to accurately discriminate similar actions, and has low recognition accuracy and poor robustness in actual environments. Firstly, in order to obtain Channel State Information (CSI) data that can better represent human activities, this paper studies the correlation between human activity changes and CSI amplitude information on different antennas, and finds that different antennas have different perceptions of human activities. This paper proposes a dynamic antenna selection algorithm based on maximum range. Secondly, due to the lack of fine spatiotemporal modeling of human behavior activities, and the integrated utilization of temporal and spatial correlations of human behavior, this paper proposes a method for human behavior detection and recognition based on the parallel model of LSTM and CNN. The CSI time series feature data is processed by the excellent performance of LSTM in processing time series, and the CSI amplitude and behavior data features are processed by the excellent performance of CNN in image processing, and finally the recognition results are obtained by working together. The experimental results show that the average accuracy of the method in aisle and laboratory scenarios is 96.2% and 93.8%, respectively.