Geometric filter algorithms for device-free localization using received-signal strength in wireless sensor networks
Talampas, Marc Caesar R.
Date of Issue2016
School of Electrical and Electronic Engineering
Device-free localization (DFL) is a method of determining the location of a target without requiring the target to wear a device or tag. This capability to track a device-free target is useful in applications where the target may be uncooperative and unwilling to be located and monitored. In radio frequency-based DFL systems that use received-signal strength (RSS) measurements, the changes induced by the target’s presence or motion on the RSS of the network’s links are used to infer his location. A number of RSS-based DFL algorithms have been recently proposed that can locate and track a target accurately, albeit with high computational requirements. This thesis presents new DFL algorithms that have lower computational costs while able to track a single device-free target with high accuracy. In this thesis, a new single target RSS-based DFL algorithm, referred to as the “Geometric Filter” (GF) algorithm is proposed. The GF algorithm uses simple geometric objects to represent radio links, probable target locations, and locational filters. The intersection points of line segments representing the target-affected links are used as probable locations of the device-free target. A locational filter is used to remove outlier links and points. Information about the target’s prior location and induced RSS changes are used to further refine the target location estimates. In order to perform accurate tracking in multipath-rich environments, the GF algorithm was extended further to utilize channel diversity. The “Multi-Channel Geometric Filter” (MCGF) fuses measurements of the RSS changes of each link across different frequency channels, and uses link-specific thresholds to detect the target-affected links. The measurements are then processed by a modified GF algorithm that uses estimates of the overall fade levels of intersecting links as weights to generate the target location estimates. The GF and MCGF algorithms have been evaluated using single-target tracking experiments in both indoor and outdoor environments. In these experiments, the new algorithms have been shown to outperform existing DFL algorithms in both tracking accuracy and execution time.