Robust radio tomographic imaging for localization of targets under uncertain sensor location scenario

• Published in: Digital signal processing 2023-06, Vol.137, p.104030, Article 104030
• Main Authors: Mishra, Abhijit, Sahoo, Upendra Kumar, Maiti, Subrata
• Format: Article
• Language: English
• Subjects: [formula omitted]-SRA; Radio tomographic imaging; Spatial loss field; Stochastic robust approximation; Worst-case robust approximation
• ISSN: 1051-2004; 1095-4333
• DOI: 10.1016/j.dsp.2023.104030

Object localization and tracking employing device-free localization (DFL) techniques have received much interest in wireless sensor networks (WSNs). One such DFL technique is radio tomographic imaging (RTI), which makes use of radio waves to image targets in wireless networks. RTI employs spatial loss fields (SLFs), which are maps that indicate the amount of attenuation of radio waves at every spatial point in the WSNs due to the presence of obstacles. The majority of recent RTI techniques neglect the practical problem of sensor position uncertainty while localizing targets. When an assumption relating to a known sensor position is violated, the estimation performance of SLFs is drastically reduced. In this paper, the above-mentioned problem is addressed through two novel robust approximation algorithms, i.e., worst-case robust approximation (WCRA) for RTI (WCRA-RTI) and stochastic robust approximation (SRA) (SRA-RTI). Furthermore, the novel SRA method based on two types of regularization techniques is proposed and denoted as l2-based-SRA (l2-SRA), l1-based-SRA (l1-SRA). The superiority of the proposed robust algorithms over the state-of-the-art methods is verified by the qualitative and quantitative approaches. •Sensor position uncertainty is an important real-world problem in RTI.•SLF estimation accuracy degrades with increase in sensor position uncertainty.•Uncertainty in sensor location leads to an uncertain weight matrix.•Robust approximation techniques are used to handle sensor location uncertainty.•SRA techniques for sparse and non-sparse scenarios provide best SLF estimation.