Publication: Ble-based indoor localization with machine learning
Loading...
Date
2024-08-01
Authors
Hor, Chia Wei
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Indoor localization has become a crucial aspect across various sectors, from
enhancing user experiences in retail environments to optimizing logistics in industrial
settings. As machine learning technology progresses, the efficacy of indoor
localization systems has substantially improved. Despite these advancements,
Bluetooth Low Energy (BLE)-based localization systems present ongoing
opportunities for further enhancement. This thesis introduces a Temporal
Convolutional Network (TCN) as a novel classification model for indoor localization,
striking a balance between efficiency and accuracy, thereby positioning it as an ideal
solution for indoor localization tasks. The methodology involves segmenting a 10.72
meters by 6.3 meters indoor area into a grid on the (x, y) plane with cells measuring
0.8 meters each. Received Signal Strength Indicator (RSSI) data collected from three
strategically placed BLE beacons within this grid were processed in MATLAB,
utilizing both moving average and median filters. Subsequently, the data was subjected
to machine learning analysis. Comparative evaluations of the TCN model against
traditional classifiers, including Decision Trees (DT), Support Vector Machines
(SVM), and k-Nearest Neighbors (kNN), underscore the superior performance of the
TCN. Notably, TCN outperforms the kNN model by 10% and exceeds the nearest
reference-based fingerprinting method by 37.77% in accuracy. These results affirm the
effectiveness of the proposed TCN approach in enhancing indoor localization systems.