Integration of Artificial Neural Network Regression and Principal Component Analysis for Indoor Visible Light...

Area of Science:

• Indoor positioning systems
• Artificial intelligence in localization
• Optical wireless communication

Background:

• Visible-light positioning (VLP) offers infrastructure-free indoor localization but faces challenges from light propagation, unknown LED poses, and dynamic environments.
• Conventional Received Signal Strength (RSS)-based methods struggle with optical power fluctuations and modeling imperfections.
• User mobility and obstacles further complicate accurate indoor positioning.

Purpose of the Study:

• To develop a robust and accurate indoor positioning system using visible light.
• To address the complexities of VLP in dynamic indoor environments and unknown LED poses.
• To leverage machine learning for improved VLP performance.

Main Methods:

• Simulated user movement with varied receiver positions to create realistic datasets.
• Implemented a regression-based artificial neural network (ANN) model for VLP.
• Utilized Principal Component Analysis (PCA) for dimensionality reduction to optimize ANN performance.
• Experimented with a constellation of eight LEDs and a photodiode receiver in a 12m x 18m x 6.8m room.

Main Results:

• The PCA-ANN model demonstrated significant improvements in positioning accuracy.
• Achieved low Mean Squared Error (MSE) values of 0.0062 cm (training) and 0.0456 cm (testing).
• Attained high R-squared values of 99.31% (training) and 94.74% (testing), indicating robust predictive performance.

Conclusions:

• The proposed PCA-ANN regression model effectively optimizes Visible-Light Positioning systems.
• This approach provides a feasible and reliable solution for indoor positioning services.
• PCA significantly enhances the computational efficiency and accuracy of VLP systems.