Radio frequency (RF) sensing networks are a class of wireless sensor networks (WSNs) which use RF signals to carry out tasks such as tomographic imaging, tomographic target tracking and node localization. While a wide variety of such algorithms exist, they often assume access to measurements of the baseline received signal strength (RSS) on each link, i.e, to measurements taken during some offline calibration period when no temporary obstructions are located near the nodes which form the network. However, in many cases, WSNs are designed to be deployed and used on the fly, and it can be impossible to ensure the network is empty of obstructions long enough to obtain the required calibration data. For instance, an RF sensing network could be set up around a burning building to image its interior and determine if people are trapped inside. There is no way to ask these people to first leave the area while the baseline RSS values are collected.;Thus far, no research has addressed the question of whether it is possible to estimate baseline RSS values without access to a calibration period. We propose adapting background subtraction methods from the fields of computer vision and image processing in order to estimate baseline RSS values from measurements taken while the system is online and obstructions may be present in the network. This is done by forming an analogy between the intensity of a background pixel in an image and the baseline RSS value of a link in a WSN. We also translate the concepts of temporal similarity, spatial similarity and spatial ergodicity which underlie three specific background subtraction algorithms---background subtraction with temporal background modelling, foreground-adaptive background subtraction and foreground-adaptive background subtraction with Markov modelling of change labels---to the domain of WSNs in order to use these algorithms to determine the baseline RSS.;By applying these techniques to experimental data, we show that they are capable of accurately estimating baseline RSS values in a range of different environments. We also show that these estimates are close enough to the actual values of the baseline RSS to allow for RF tomographic tracking to be carried out without the need to resort to a calibration period.
展开▼
