Multiple People Identification Through Walls Using Off-the-Shelf WiFi

Abstract

In this article, we are interested in through-wall gait-based identification of multiple people who are simultaneously walking in an area, using only the WiFi magnitude measurements of a small number of transceivers. This is a considerably challenging problem as the gait signatures of the walking people are mixed up in the WiFi measurements. In order to solve this problem, we propose a novel multidimensional framework, spanning time, frequency, and space domains, that can separate the signal reflected from each walking person and extract its corresponding gait content, in order to identify multiple people through walls. To the best of our knowledge, this is the first time that WiFi signals can identify multiple people in an area. We extensively validate our proposed system with 92 test experiments conducted in four different areas, where the WiFi transceivers are placed behind walls, and where two or three people (randomly selected from a pool of six test subjects) are walking in the area. Our system achieves an overall average accuracy of 82% in correctly identifying whether a person walking in the test experiment (referred to as a query) is the same as a candidate person, based on 6404 query-candidate test pairs. It is noteworthy that none of the test subjects/areas has been seen in the training phase.