Object tracking system #2 (NN model development)

Abstract

Bluetooth Low Energy (BLE) has emerged as a promising solution for asset tracking, addressing the limitations of traditional methods such as GPS and Wi-Fi. GPS signals struggle to penetrate enclosed spaces, making indoor tracking unreliable. Wi-Fi-based localization, while more suitable for indoor environments, typically suffers from lower accuracy (5–10 meters) due to complex RF propagation while also requiring high setup costs in areas that do not have Wi-Fi access. BLE offers a potential alternative with lower power consumption and cost-effective deployment. However, despite its advantages, BLE signals are not yet fully understood, with certain properties still unexplored. These challenges necessitate advanced techniques to improve the accuracy of BLE-based localization. This Study evaluates the performance of various deep learning techniques in improving the accuracy and reliability of BLE-based localization systems. A novel data augmentation strategy is introduced to expand the training dataset and mitigate overfitting. The proposed methods include implementations of Deep Neural Networks (DNNs) and Convolutional Neural Networks (CNNs). We evaluate both grid-based classification and regression-based localization approaches. Our approach includes fingerprinting for feature extraction and transfer learning during deployment to a new location. Additionally, a real-time asset tracking web application is developed to demonstrate the usability of the trained models in practical environments. Preliminary results indicate that these techniques significantly enhance BLE-based tracking precision, providing a reliable model for real-time indoor asset tracking. Future work may explore the impact of different environmental variables and leverage Recurrent Neural Networks (RNN) or Transformers to exploit any temporal dependencies.