Fingerprinting-based indoor localization: A Deep Learning Approach

Abstract

Achieving accurate indoor localization is of paramount importance for numerous applications, including asset tracking, navigation, and context-aware services. In this research, we propose a design and an implementation of a deep Convolutional Neural Network (CNN) classification model for indoor localization. The model is trained and tested using a rich labeled dataset encompassing four different indoor environments sharing a common characteristic of being located on the same floor within the same building. Each environment is characterized by varying levels of clutter: highly cluttered, medium cluttered, and low cluttered open spaces. The experimental results demonstrate a remarkable increase in localization accuracy across all environments. The average accuracy achieved by the deep CNN classification model exceeds 99%. This impressive performance highlights the model's ability to effectively distinguish and classify objects in indoor environments that exhibit varying degrees of clutter. The proposed model holds great promise for applications that rely on precise indoor localization, showcasing its potential to meet the demands of real-world scenarios.