Visible light positioning : a machine learning approach

Visible light positioning (VLP) systems have experienced substantial revolutionary progress over the past year because they can offer great positioning accuracy without needing any additional infrastructure, as conventional radio-frequency (RF)-based systems. Received signal strength (RSS)-based VLP systems are a promising approach to many indoor positioning estimation problems, but still suffer from difficulty in providing high accuracy and reliability. A potential solution to these challenges is to combine VLP systems, and machine learning (ML) approaches to enhance the position prediction accuracy in two-dimensional (2-D) spaces, or more complex problems. In this paper, we propose a ML approach to accurately predict the 2-D indoor position of a mobile receiver (eg. an automated guided vehicles-AGV), based on the measured RSS values of 4 photodiodes (PDs) forming a star architecture. We examine and evaluate the performance of different ML learners applied to the above-described problem. The proposed ML and Neural Network (NN) methods exhibit great accuracy results in predicting the 2-D coordinates of a PD-based receiver.Visible light positioning (VLP) systems have experienced substantial revolutionary progress over the past year because they can offer great positioning accuracy without needing any additional infrastructure, as conventional radio-frequency (RF)-based systems. Received signal strength (RSS)-based VLP systems are a promising approach to many indoor positioning estimation problems, but still suffer from difficulty in providing high accuracy and reliability. A potential solution to these challenges is to combine VLP systems, and machine learning (ML) approaches to enhance the position prediction accuracy in two-dimensional (2-D) spaces, or more complex problems. In this paper, we propose a ML approach to accurately predict the 2-D indoor position of a mobile receiver (eg. an automated guided vehicles-AGV), based on the measured RSS values of 4 photodiodes (PDs) forming a star architecture. We examine and evaluate the performance of different ML learners applied to the above-described problem. The proposed ML and Neural Network (NN) methods exhibit great accuracy results in predicting the 2-D coordinates of a PD-based receiver.C