Visible light channel modeling and application in underground mines based on transformer point clouds optimization

Abstract

In the unique environment of mines, the low sensitivity of point cloud acquisition cameras often results in incomplete wall point cloud data. Direct application of this data to channel models can result in significant positioning errors. To address this issue, this paper proposes a DRC-PoinTr deep 3D point cloud optimization model that integrates denoising, registration, and completion (denoising, registration and completion, DRC). To evaluate the impact of wall integrity on visible light positioning, this study creates a wall point cloud dataset that includes real application environments. The DRC-PoinTr model was employed to train the dataset, optimizing the wall point cloud data collected in real-world environments. Then, the triangular mesh of the wall's reflective surface was generated using the Delaunay triangulation approach, and the triangulated wall was applied to the visible light channel model to simulate power distribution under different conditions. Subsequently, the impact of wall optimization on visible light positioning was analyzed. The results showed that in a simulated space of 3 m × 4 m × 2 m, the average received power of the channel model after wall optimization increased by more than 60% compared to the unoptimized model. Furthermore, visible light positioning simulation experiments were conducted based on the optimized channel model. The results demonstrated that the optimized wall channel model achieved a root mean square positioning error of approximately 2 cm, satisfying the positioning accuracy requirements for underground mining environments. Finally, we performed visible light positioning experiments in a real mining environment, demonstrating the process of applying actual measured optical power data for positioning.