LEDPOS: Indoor Visible Light Positioning Based on LED as Sensor and Machine Learning
Christian Fragner, Christian Krutzler, Andreas P. Weiss, Erich Leitgeb
Abstract
Accurate indoor positioning is becoming increasingly important, especially in highly automated industrial environments with robots. In addition, LED-based lighting is also being used more and more frequently in such application fields. In the present work, the possibility to exploit the LED lighting infrastructure with a novel approach for implementing an accurate indoor positioning system is investigated. For this purpose, a demonstrator luminaire LEDPOS is proposed and evaluated that combines visible light sensing based on backscattered reflections to accurately estimate the two-dimensional position of a retroreflective foil at the floor while providing simultaneously an unimpaired room illumination. In particular, the same LED elements are shared for illumination and for the sensing functionality. Furthermore, the algorithm for data evaluation and position determination is based on a machine learning approach that is implemented on the edge in the luminaire. Thus, the presented approach allows for a simple and cost-efficient implementation in different applications. The experimental characterization of the LEDPOS demonstrator in a real-world scenario shows that a very good positioning accuracy can be achieved, in which the average error for the two-dimensional position of the retroreflective foil within an area of 0.64 m <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> remains in the range of 3 cm.