Robust Near-Field 3D Localization of an Unaligned Single-Coil Agent Using Unobtrusive Anchors

TL;DR

This paper presents a robust method for 3D indoor localization using magnetic near-field measurements from flat coils, achieving near-theoretical accuracy with a fast, initialization-insensitive algorithm suitable for real-time applications.

Contribution

It introduces a novel weighted least squares algorithm that effectively handles unknown orientation and scales anchor observations, enabling accurate and robust 3D localization.

Findings

Achieves cm-level accuracy in a 10x10x3 m room with 12 anchors.

The proposed WLS algorithm is insensitive to initialization and scales well.

Localization accuracy near the theoretical limit in up to 98% of cases.

Abstract

The magnetic near-field provides a suitable means for indoor localization, due to its insensitivity to the environment and strong spatial gradients. We consider indoor localization setups consisting of flat coils, allowing for convenient integration of the agent coil into a mobile device (e.g., a smart phone or wristband) and flush mounting of the anchor coils to walls. In order to study such setups systematically, we first express the Cram\'er-Rao lower bound (CRLB) on the position error for unknown orientation and evaluate its distribution within a square room of variable size, using 15 x 10cm anchor coils and a commercial NFC antenna at the agent. Thereby, we find cm-accuracy being achievable in a room of 10 x 10 x 3 meters with 12 flat wall-mounted anchors and with 10mW used for the generation of magnetic fields. Practically achieving such estimation performance is, however, difficult because of the non-convex 5D likelihood function. To that end, we propose a fast and accurate weighted least squares (WLS) algorithm which is insensitive to initialization. This is enabled by effectively eliminating the orientation nuisance parameter in a rigorous fashion and scaling the individual anchor observations, leading to a smoothed 3D cost function. Using WLS estimates to initialize a maximum-likelihood (ML) solver yields accuracy near the theoretical limit in up to 98% of cases, thus enabling robust indoor localization with unobtrusive infrastructure, with a computational efficiency suitable for real-time processing.