VLC localization: deep learning models by Kalman filter algorithm combined with RSS

Abstract In this paper, a new framework is presented for indoor visible light communication (VLC) system, based on Yolo v3, EfficientNetB3, and DenseNet121 deep learning (DL) models, as well an optimization strategy. The proposed consists of two steps: data collecting DL model training. To start, acquired using MATLAB Kalman Filtering (KF) with averaging approaches. Second, the received signal strength (RSS) employed models input, Cartesian coordinates output. RSS approach combined KF algorithm are used in suggested framework. This work introduces impacts Non-Line-of-Sight (NLoS) initial reflection Line-of-Sight (LoS) three mentioned models. Furthermore, we Bayesian automatic hyper-parameter (HP) to increase system efficiency reduce positioning error obtained results show that outperform existing HP-RSS-KF-LoS-DL terms localization when compared traditional signal-based techniques. Many performance indicators considered evaluate resiliency, including accuracy (ACC), area under curve (AUC), sensitivity (Se), precision (Pr), F1-score, root mean square (RMSE), training, testing time. generated trained Python software Kaggle Notebook GPU cloud (2 CPU cores 13 GB RAM). achieved are: 99.99% ACC, 99.98% AUC, 98.88% Se, 98.98% Pr, 99.97% 0.112 cm RMSE, 0.29 s could be easily deployed autonomous applications, analysis experimental data. Several applications can depending enhancing VLC military systems, underwater systems like hospitals, hotels, libraries malls.