An Efficient CSI-Based Pedestrian Monitoring Approach via Single Pair of WiFi Transceivers

Pedestrian monitoring is a very important issue in many sensitive areas. Traditional technologies mainly include computer vision, infrared imaging, WiFi sniffing, etc. Nevertheless, these technologies cannot simultaneously satisfy low cost, user privacy and high reliability requirements. In view of this, this work aims at proposing a method based on CSI (Channel State Information) to enable pedestrian monitoring by leveraging a single pair of WiFi transceivers, which can work under different temperatures and light conditions as well as in a non-intrusive way. First, we pre-process the raw CSI to sift the interested components. Then, a two-stage clustering method is proposed to counteract the multipath effect, where the CSI pattern for pedestrian activities is learned in the offline phase. During the online phase, a Pedestrian Pass Detection Algorithm (PPDA) is proposed for pass detection. Further, we propose a Pass Direction Recognition Algorithm (PDRA) for direction recognition, by calculating the time of pass Line Of Sight (LOS) and the pass direction indicator. Finally, we implement the system prototype and conduct a series of real-word experiments, and the results conclusively demonstrate the feasibility and efficiency of the proposed methods.