Optimization of Visible Light Positioning in Industrial Applications using Machine Learning

This paper investigates the error performance of visible Light Positioning (VLP) systems for 3D indoor drone localization using machine learning algorithms. Received Signal Strength is used to track the position of the drone and different smoky channel conditions to emulate an industrial environment. VLP systems utilize visible light communication (VLC) and indoor positioning, providing a low-cost and interference-free solution for precise drone localization. Machine learning (ML) based artificial neural network (ANN) is used to trained on diverse datasets and correlations between received signal strength (RSS) measurements and position errors. The results demonstrate that ML enables accurate real-time drone position estimation, compensating for atmospheric attenuation. The trained models achieve significantly improved localization accuracy and capturing non-linear relationships between input features and drone location. Furthermore, machine learning algorithms extract relevant features, reducing the impact of noise and atmospheric attenuations. ML process enhances the VLP system’s robustness, resulting in remarkable localization accuracy improvements compared to the attenuated path with average error values from 21.9 cm to 5.9 cm. The trained ML models achieve RMSE values of 0.044772 and 0.067523, respectively, with high R-squared values of 0.999. Furthermore the error histogram analysis confirms accurate drone location estimation, even in the presence of atmospheric attenuations