Long Short-Term Memory Fully Convolutional Network-Based Passing-Object Detection and Its Application to WLAN Sensing

Abstract

Wireless sensing technology is the cornerstone of realizing an intelligent environment. As an application of wireless sensing, the research significance of passing-object detection has been verified. This paper proposes a sensing method for passing-object detection, in which the channel state information (CSI) is combined with long short-term memory fully convolutional network (LSTM-FCN). In the method, the CSI is firstly extracted from the multiple antennas and multiple subcarriers, and the transmission waveform is based on the frame structure of the IEEE 802.11ac protocol. Then, the principal components analysis (PCA) is applied to reduce the dimension of CSI data. At the model train stage, the LSTM-FCN is applied to train the first principal component information of CSI data. The proposed method is implemented and validated with extensive experiments in indoor corridor environments, and the verified accuracy of the model exceeds 96%.