Feasibility and limits of wi-fi imaging

TLDR: This work implements a prototype wireless receiver using USRP-N210s at 2.4 GHz and demonstrates that it can image objects such as leather couches and metallic shapes in line-of-sight and non-line ofsight scenarios and can localize static human subjects and metallic objects with a median accuracy of 26 and 15 cm respectively.

Abstract: We explore the feasibility of achieving computational imaging using Wi-Fi signals. To achieve this, we leverage multi-path propagation that results in wireless signals bouncing off of objects before arriving at the receiver. These reflections effectively light up the objects, which we use to perform imaging. Our algorithms separate the multi-path reflections from different objects into an image. They can also extract depth information where objects in the same direction, but at different distances to the receiver, can be identified. We implement a prototype wireless receiver using USRP-N210s at 2.4 GHz and demonstrate that it can image objects such as leather couches and metallic shapes in line-of-sight and non-line-of-sight scenarios. We also demonstrate proof-of-concept applications including localization of static humans and objects, without the need for tagging them with RF devices. Our results show that we can localize static human subjects and metallic objects with a median accuracy of 26 and 15 cm respectively. Finally, we discuss the limits of our Wi-Fi based approach to imaging.