Target Localization by Unlabeled Range Measurements

Abstract

In this paper, the unlabeled target localization problem in a wireless sensor network is addressed. Sensors of the network are deployed in a two-dimensional surveyed area and measure their distance to a target. A central unit processes the data delivered by the sensors and is tasked to infer the position of a target, but it must do so without knowing the association between the received data and the identity of the delivering sensors. This setting of unlabeled inference is attracting much attention in recent years due to its relevance to emerging applications involving big data. The present article exploits the underlying geometrical structure of the inference problem to develop two unlabeled localization approaches based on range measurements, referred to as partition and intersection points methods, respectively. The design of these methods is based on analytical results while their performance assessment is conducted by computer experiments. We find that the partition method should be preferred in the regime of low signal-to-noise ratio and networks consisting of a small number of sensors, while the intersection points method is more suitable to solve the target localization problem with large networks.