Multi-view WiFi imaging

Abstract

Listen

Translate article

Data driven models outperform their model driven counterparts, and model driven models are easier to explain and extend. Bridging the gap between these approaches can benefit from the desirable features on both ends. In this respect, a data driven model proposed on the foundations of model driven methods in wireless imaging can extend the single-view imaging problem to the multi-view scenario. Accordingly, we propose two multi-view approaches based on the plenoptic model. The plenoptic model is rooted in computational imaging and captures what the observer perceives. We establish that the wireless imaging problem can be formulated as an association problem between the plenoptic function in the visible light and the radio frequency spectrum. We then develop two Deep Neural Networks based on the expressed association problem. Both models can generate three camera views from WiFi channel state information of a single receiver. These models are the first solutions to extend WiFi imaging to multi-view settings. We demonstrate our method’s performance and illustrate that it can be an effective stand-alone tool or to augment visible light frameworks.