A visible light positioning system for coal mine personnel based on convolutional recurrent neural network

Abstract

Visible Light Positioning (VLP) technology exhibits promising potential for underground coal mine positioning. While numerous studies have explored VLP in indoor environments, attention to VLP systems for underground mines still needs to be improved. In this paper, based on the traditional indoor visible light communication channel model, the effects of unique features on VLP, such as irregular walls, inclined and rotating receiving ends, and coal dust particles in underground mines, are considered and modeled in detail. In order to improve the accuracy and reliability of personnel VLP in complex mine environments, this paper proposes a deep-learning-based VLP system for underground personnel in coal mines, which consists of two LEDs and four photodetectors (PDs), where the four PDs are mounted on miners' helmets, in conjunction with the actual mine environment. Due to environmental factors and noise, there are apparent spatio-temporal characteristics in the received signal strength of the PDs. Therefore, the system adopts a convolutional recurrent neural network (CRNN) for feature extraction. In the experimental phase, the method's average localization error was 11.24 cm, with 90% of the localization errors within 20.4 cm. The experimental results show that the system has high positioning accuracy and meets the requirements of personnel positioning in underground mines.