| Description |
Device-free localization (DFL) and tracking services are important components in security, emergency response, home and building automation, and assisted living applications where an action is taken based on a person's location. In this dissertation, we develop new methods and models to enable and improve DFL in a variety of radio frequency sensor network configurations. In the first contribution of this work, we develop a linear regression and line stabbing method which use a history of line crossing measurements to estimate the track of a person walking through a wireless network. Our methods provide an alternative approach to DFL in wireless networks where the number of nodes that can communicate with each other in a wireless network is limited and traditional DFL methods are ill-suited. We then present new methods that enable through-wall DFL when nodes in the network are in motion. We demonstrate that we can detect when a person crosses between ultra-wideband radios in motion based on changes in the energy contained in the first few nanoseconds of a measured channel impulse response. Through experimental testing, we show how our methods can localize a person through walls with transceivers in motion. Next, we develop new algorithms to localize boundary crossings when a person crosses between multiple nodes simultaneously. We experimentally evaluate our algorithms with received signal strength (RSS) measurements collected from a row of radio frequency (RF) nodes placed along a boundary and show that our algorithms achieve orders of magnitude better localization classification than baseline DFL methods. We then present a way to improve the models used in through-wall radio tomographic imaging with E-shaped patch antennas we develop and fabricate which remain tuned even when placed against a dielectric. Through experimentation, we demonstrate the E-shaped patch antennas lower localization error by 44% compared with omnidirectional and microstrip patch antennas. In our final contribution, we develop a new mixture model that relates a link's RSS as a function of a person's location in a wireless network. We develop new localization methods that compute the probabilities of a person occupying a location based on our mixture model. Our methods continuously recalibrate the model to achieve a low localization error even in changing environments. |
