A semi-parametric model for target localization in distributed systems

TL;DR

This paper introduces a nonparametric, tuning-free estimator called SSCE for target localization in distributed sensor systems, demonstrating its consistency, asymptotic normality, and practical effectiveness through simulations and real data application.

Contribution

It proposes a novel nonparametric target localization method that is tuning-free, consistent, and has a Gaussian limit distribution, advancing existing parametric approaches.

Findings

SSCE is $ oot{n}$ consistent and asymptotically normal.

The method performs well in simulations and real surveillance data.

It provides a practical alternative to parametric models for target localization.

Abstract

Distributed systems serve as a key technological infrastructure for monitoring diverse systems across space and time. Examples of their widespread applications include: precision agriculture, surveillance, ecosystem and physical infrastructure monitoring, animal behavior and tracking, disaster response and recovery to name a few. Such systems comprise of a large number of sensor devices at fixed locations, wherein each individual sensor obtains measurements that are subsequently fused and processed at a central processing node. A key problem for such systems is to detect targets and identify their locations, for which a large body of literature has been developed focusing primarily on employing parametric models for signal attenuation from target to device. In this paper, we adopt a nonparametric approach that only assumes that the signal is nonincreasing as function of the distance between the sensor and the target. We propose a simple tuning parameter free estimator for the target location, namely, the simple score estimator (SSCE). We show that the SSCE is $\sqrt{n}$ consistent and has a Gaussian limit distribution which can be used to construct asymptotic confidence regions for the location of the target. We study the performance of the SSCE through extensive simulations, and finally demonstrate an application to target detection in a video surveillance data set.