One for More: Sparse Group Lasso for Multitarget Device-Free Localization With Single-Target Fingerprint Database Under RFID System

Device-free localization (DFL) has become a hot spot in the field of indoor localization in recent years since DFL does not need the target to be localized to carry a device. Fingerprinting is a data-driven approach and does not require line-of-sight propagation. However, the problem is that the complexity of the fingerprint database exponentially increases with the number of targets. This makes the fingerprinting-based multitarget DFL a complex problem. In this article, we propose a sparse group lasso-based DFL (SGL-DFL) scheme to realize multitarget DFL with a traditional single-target fingerprint database. SGL-DFL contains two phases: in the phase of fuzzy coarse localization, we use a weighted voting method to generate a heatmap. Taking the local maximums as group centers, we propose a simultaneous outlier elimination and grouping scheme to obtain several groups of possible reference fingerprints. The average hot value of each group is used to evaluate the likelihood that the target exists in each group. In the phase of fine localization, the multitarget DFL is formulated as a multicomponent optimization technique in the form of sparse group lasso. Compared with existing fingerprinting-based multitarget DFL, SGL-DFL not only considers the collective influence of multiple targets on wireless signals but also considers the fingerprint matching degree of each target. Moreover, SGL-DFL does not need to know the location of the communication node. Experimental evaluations in a
$4\times8$
m area in an indoor warehouse environment achieved the mean distance error of 0.85, 1.40, and 1.45 m for two-, three-, and four-target localizations, respectively.