Measurement-Based Training Method for Enhancing Device-Free Localization Accuracy

Abstract

Device-Free Localization (DFL) is a promising approach for smart environments and disaster response because it enables non-intrusive tracking without requiring users to carry radio devices. Radio Tomographic Imaging (RTI) enables accurate DFL by exploiting shadowing effects induced by targets on wireless links. This method models link paths between anchor nodes using a voxel-based map and estimates target locations by superimposing shadowed paths without prior environmental measurements or fingerprinting. However, robust performance depends strongly on the selection of model parameters. Because determining these parameters requires extensive calibration measurement, efficient procedures are essential for practical use. To address this issue, this letter proposes a measurement-based training strategy for Bayesian-optimized multipath-RTI and investigates how different training-point configurations affect localization performance. Experimental evaluations at 25 indoor positions demonstrate that the proposed strategy achieves stable and generalizable localization performance across different spatial conditions with reduced calibration effort.