A generalized thresholding algorithm with dimension reduction for device-free localization in IoT

doi:10.1007/s10489-022-03925-2

2022-08-05

PRE

DOI：10.1007/s10489-022-03925-2

原文链接

原文求助

摘要

En 中文

Location information is one of the most important factors for many location-based services (LBSs) in the Internet of Things (IoT). Device-free localization (DFL) has received more attention as it achieves localization without attaching any electronic device to the target. DFL can be applied to many special scenarios, such as monitoring the elderly living alone, health care of inpatients, and emergency rescue. In applications based on traditional localization methods, the numerous receive signal strength (RSS) measurements are collected from wireless sensor networks (WSNs) comprised of sensor pairs to construct the atoms of learning dictionaries. With recovery algorithms, solutions can be obtained from undetermined equations using learning dictionaries, which can be mapped to the position index of the target to estimate the accurate coordinates. However, the numerous RSS data produced by WSN sensor generate high-dimensional learning dictionaries that cost the sparse recovery algorithm more iterative computation time to derive the target location and more space for data storage, thus affecting the real-time DFL performance. In this paper, we propose a data dimension reduction method based on the generalized iterative thresholding algorithm for DFL. Firstly, we reduced the column and row dimensions of the dictionary, respectively, via principal components analysis (PCA). Then, the dimension of the observed vector was reduced correspondingly. Finally, the new underdetermined equation was solved via sparse coding with an iterative p-thresholding algorithm in signal subspace, and the target location was estimated accurately. Experiments on public datasets demonstrated that the proposed method outperforms the current alternatives by improving the computation efficiency of DFL systems and taking less time to locate the target, implying its good applicability to IoT scenarios with high real-time requirements.

Keyword:

Device-free localization

p generalized thresholding function

Dimension reduction

Principal component analysis