Machine Learning Based Visible Light Indoor Positioning With Single-LED and Single Rotatable Photo Detector

Abstract

In recent years, visible light positioning (VLP) systems have attracted considerable attention because they do not require additional infrastructures. However, most existing researches on the VLP ignore the impact of wall diffuse reflection, which can lead to the dramatical decrease of the position accuracy performance near the indoor walls and corners. In this paper, we design an indoor VLP system with one single light-emitting diode (LED) and a rotatable photodetector (PD), and then propose an indoor VLP algorithm based on machine learning (ML) methods with concern for the indoor reflection of the optical propagation. The proposed positioning process is implemented via two stages: area classification and positioning. During the area classification stage, by using the random forest (RF) algorithm, the entire room is divided into one interior area and four wall or corner zones. In the interior area, the rotatable PD is directly used to determine the target location. In the four wall or corner zones, a hybrid positioning algorithm based on the extreme learning machine (ELM) and the density-based spatial clustering of applications with noise (DBSCAN) is developed to improve localization accuracy near the indoor walls and corners. Simulation results show that by using the proposed indoor VLP system with the rotatable PD and the hybrid algorithm, the maximum and averaged positioning errors of wall or corner zones drop from 137.96 cm and 15.63 cm, to 38.34 cm and 1.43 cm, respectively, and the averaged positioning error of the whole room decreases from 11.97 cm to 1.74 cm.

Keywords

• Density-based spatial clustering of applications with noise (DBSCAN)
• extreme learning machine (ELM)
• Internet of things (IoT)
• random forest (RF)
• rotatable PD
• single LED