Machine Learning aided Precise Indoor Positioning

TL;DR

This paper presents a UWB and machine learning-based indoor positioning system that reduces measurement effort through a mathematical data generation strategy, achieving less than 150 mm average error with the best model.

Contribution

It introduces a novel mathematical approach to generate data, decreasing measurement requirements in fingerprint-based indoor localization systems.

Findings

Average error less than 150 mm at most locations

Four models compared for performance

Data generation reduces measurement effort

Abstract

This study describes a UWB and Machine Learning (ML)-based indoor positioning system. We propose a simple mathematical strategy to create data to reduce the job of measurements for fingerprint-based indoor localization systems. A considerable number of measurements can be avoided this way. The paper compares and contrasts the performance of four distinct models. Most test locations' average error may be reduced to less than 150 mm using the best model.