A Multiple Wi-Fi Sensors Assisted Human Activity Recognition Scheme for Smart Home

With the development of wireless communication technology, wireless signals have been expanded from mobile communication to behavioral sensing. In particular, human activity recognition (HAR) relying on Wi-Fi signals has attracted increasing attention and demonstrated its great potential in the field of smart healthcare. However, most HAR solutions fail to capture the full scope of the relationship between Wi-Fi signals and human activities. To address this issue, we develop a novel depthwise separable convolution neural network (DSCNN)-based HAR by using channel state information (CSI) of multiple Wi-Fi access points (APs) sensors. To this end, we first introduce an activity-related information enhancement (ARIE) strategy to extract useful information from the CSI and mitigate background noises. Then, we design a multiview CSI fusion (MVCF) approach to calculate key features by aggregating the CSI measurements from all Wi-Fi APs. With this strategy, the feature describes the data themselves more comprehensively than a single view individually. Finally, a DSCNN is used to capture a reliable mapping between the key feature and daily activities, without a scenario-specific calibration. We design and test an HAR prototype on commodity Wi-Fi devices and perform experiments in typical indoor environments. Experimental results confirm that the proposed scheme can achieve accurate and robust HAR compared with existing ones.