Fig. 1. Indoor visible light imaging position sensing system

Fig. 2. Schematic diagrams of 3×3 conventional convolution and deformable convolution sampling method.(a) Collected visible light image; (b) conventional convolution sampling point arrangement; (c) deformable convolution sampling point arrangement

Fig. 3. Deformable convolution feature extraction process

Fig. 4. Feature extraction of visible light image

Fig. 5. Indoor visible light positioning process

Fig. 6. Construction of experimental environment

Fig. 7. Visible light images captured

Fig. 8. Schematic diagram of mobile terminal movement process

Fig. 9. Mobile terminal rotation

Fig. 10. Mobile terminal tilt

Fig. 11. Comparison of matching results between this algorithm and SuperPoint algorithm. (a) Algorithm of this study; (b) SuperPoint algorithm

Fig. 12. Comparison of matching results of the visible light image in the central region by rotating 30°. (a) Algorithm of this study; (b) SuperPoint algorithm

Fig. 13. Comparison of matching results of the edge region visible light image by rotating 30°. (a) Algorithm of this study; (b) SuperPoint algorithm

Fig. 14. Comparison of matching results under different transformations of visible light images. (a) Rotation (left rotation) at different angles; (b) tilt (left tilt) at different angles

Fig. 15. Positioning distribution of the actual position and the predicted position at h=0

Fig. 16. Positioning distribution of the actual position and the predicted position at h=0.75 m

Fig. 17. Positioning distribution of the actual position and the predicted position at h=1.50 m

Fig. 18. Error distribution diagrams at heights of h=0, 0.75, 1.50 m. (a) h=0; (b) h=0.75 m; (c) h=1.50 m

Fig. 19. Attention mechanism enhances image matching

Fig. 20. Influence of different light environments on indoor visible light positioning

Fig. 21. Network time consumption of different number of feature points

Fig. 22. Training results on different partitioned datasets. (a) 3∶7 split the dataset; (b) 5∶5 split the dataset; (c) 7∶3 split the dataset