Abstract
- Geomagnetic positioning, in principle, enables infrastructure-free indoor localization—provided that spatial variations in the magnetic field can be distinguished with sufficient resolution. Neural networks represent a powerful model class for recognizing complex spatio-temporal patterns in the magnetic signal landscape.This paper presents a systematic survey of research that employs neural algorithms to process geomagnetic measurements for the purpose of positioning. Feature representations, model architectures, ground-truth strategies, and multimodal fusion approaches are compared. A comprehensive tabular overview allows direct comparison of the employed data sets, feature sets, algorithms, and the operationalization of both the reference and localization phases.