Optical Camera Communications and Machine Learning for Indoor Visible Light Positioning

Abstract

The potential of VLP is increasing with the rise of indoor mobile machine applications. In this paper, a 3D indoor VLP system based on machine learning and optical camera communications is presented. The system uses electronically controlled LED luminaires as reference points and a rolling shutter CMOS sensor as the receiver. The LED luminaires are modulated using On-Off Keying with unique frequencies. YOLOv5 is used for classification and estimation of the position of each LED luminaire in the image. The pose of the receiver is estimated using a perspective-n-point algorithm. The system was validated using a real-world sized setup containing eight LED luminaires, and achieved an average positioning error of 3.5 cm. The average time to compute the camera pose is approximately 52 ms, which makes it suitable for real-time positioning. To the best of our knowledge, this is the first application of the YOLOv5 algorithm in the field of VLP for indoor environments.

Keywords

• visible light positioning
• indoor positioning system
• three-dimensional positioning
• cmos image sensor
• rolling shutter
• perspective-n-point
• machine learning
• convolutional neural network
• yolo