A Survey of Application of Machine Learning in Wireless Indoor Positioning Systems

TL;DR

This survey reviews machine learning techniques applied to wireless indoor positioning, analyzing various data inputs, methods, and datasets to evaluate accuracy, scalability, and complexity for diverse indoor localization applications.

Contribution

It provides a comprehensive comparison of ML-based indoor positioning methods, highlighting their advantages, limitations, and suitability for different indoor localization scenarios.

Findings

RSSI-based fingerprinting achieves high accuracy in specific environments

Time-based approaches offer better scalability for large areas

ML methods vary in complexity and computational requirements

Abstract

Indoor human positioning has become increasingly important for applications such as health monitoring, breath monitoring, human identification, safety and rescue operations, and security surveillance. However, achieving robust indoor human positioning remains challenging due to various constraints. Numerous attempts have been made in the literature to develop efficient indoor positioning systems (IPSs), with a growing focus on machine learning (ML) based techniques. This paper aims to compare and analyze current ML-based wireless techniques and approaches for indoor positioning, providing a comprehensive review of enabling technologies for human detection, positioning, and activity recognition. The study explores different input measurement data, including RSSI, TDOA, etc., for various IPSs. Key positioning techniques such as RSSI-based fingerprinting, Angle-based, and Time-based approaches are examined in conjunction with various ML methods. The survey compares the positioning accuracy, scalability, and algorithm complexity, with the goal of determining the suitable technology in various services. Finally, the paper compares distinct datasets focused on indoor localization, which have been published using diverse technologies. Overall, the paper presents a comprehensive comparison of existing techniques and localization models.