Passive through-wall counting of people walking using WiFi beamforming reports

This article develops a system for through-wall counting of people walking in a room, based purely on passive reception of the WiFi signals that are generated by devices in that room. We use WiFi compressed beamforming reports, collected using a sniffer node located outside the room. We propose a 2-D discrete Fourier transform (2D DFT) approach for feature extraction. As such, we formulate the counting problem as a multiclass image classification problem. Our proposed system achieves accuracies of 100%, 97.8%, 78.3%, and 93.9% in field trials with zero, one, two, and three people walking inside a room, respectively, even for rooms that were not part of the training set.