Pano2RSSI: Generation of RSSI maps for a room environment from a single panoramic image

We consider the feasibility of predicting received signal strength indicator (RSSI) map for a room environment from a single 360° RGB panoramic image of the room using deep learning (DL). We are motivated by significant applications in rapid and automated deployment of indoor wireless sensor networks. In our knowledge, this is the first work that addresses the feasibility of RSSI prediction from visual input using DL. As a first step towards this, we propose a system, Pano2RSSI, that consists of two deep neural network (DNN) based subsystems in cascade. A single RGB panoramic image of the room environment is fed as input to the first subsystem (Pano2Layout). Pano2Layout predicts the layout of the room as well as detects objects and their sizes within. This layout information is the input to the second subsystem (RSSI-Net) which predicts a 2D RSSI map for a given 2D transmitter location within the room. In this initial proposal of the system, RSSI-Net assumes that some parameters about the wireless propagation environment are fixed (such as antenna gains, path loss exponent, material permittivities.) We illustrate the end-to-end performance of Pano2RSSI and identify several challenges and possible improvements for this problem.