TW-See: Human Activity Recognition Through the Wall With Commodity Wi-Fi Devices

Device-free passive human activity recognition plays an important role in many applications, such as smart homes, identification, health care, etc. However, existing human activity recognition systems either require a dedicated device or do not meet the scenarios of the signals through the wall. To address this challenge, we present TW-See, a device-free passive human activity recognition system with Wi-Fi signals, which does not require any dedicated device and meets the scenarios of the signals through the wall. TW-See mainly exploits two key techniques to recognize different human activities. First, we propose an opposite robust PCA (Or-PCA) approach to obtain the correlation between human activity and its resulting changes in channel state information values that can eliminate the influence of the background environment on correlation extraction. Second, we propose a normalized variance sliding windows algorithm to segment the human action time from the Or-PCA waveforms, which can distinguish the true or false of the start and end times of human actions. Furthermore, we implemented TW-See with commodity Wi-Fi devices and evaluated it in several different environments. Experimental results show that TW-See achieves an average accuracy of 94.46% when the signals pass through the concrete wall.