Decimeter level passive tracking with wifi

Abstract

Pioneer approaches for WiFi-based sensing usually employ learning-based techniques to seek appropriate statistical features, but do not support precise tracking without prior training. Thus to advance passive sensing, the ability to track fine-grained human mobility information acts as a key enabler. In this paper, we proposed Widar, a WiFi-based tracking system that simultaneously estimates human's moving velocity (both speed and direction) and locations at decimeter level. Instead of applying statistical learning techniques, Widar builds a theoretical model that geometrically quantifies the relationships between CSI dynamics and user's location and velocity. On this basis, we propose novel techniques to identify PLCR components related to human movements from noisy CSIs and then derive a user's locations in addition to velocities. We implement Widar on commercial WiFi devices and validate its performance in real environments. Our results show that Widar achieves decimeter-level accuracy, with a median location error of 24cm given initial positions and 36cm without them and a mean relative velocity error of 11%.