Application of Visible Light Positioning System Based on Fingerprint Correction and Network Fusion Algorithms

Jianyong Yu; Xiaoli Hu; Qian Wang; Ling Qin; Fengying Wang; Xinchao Kou

doi:10.3788/AOS250821

Acta Optica Sinica, Volume. 45, Issue 16, 1606005(2025)

Application of Visible Light Positioning System Based on Fingerprint Correction and Network Fusion Algorithms

Jianyong Yu¹, Xiaoli Hu^2、*, Qian Wang¹, Ling Qin^1、**, Fengying Wang³, and Xinchao Kou¹

¹School of Digital and Intelligent Industry, Inner Mongolia University of Science and Technology, Baotou 014010, Inner Mongolia , China

²School of Automation and Electrical Engineering, Inner Mongolia University of Science and Technology, Baotou 014010, Inner Mongolia , China

³Engineering Training Center, Inner Mongolia University of Science and Technology, Baotou 014010, Inner Mongolia , China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(18)

Fig. 1. Real environment channel model

Download full size

Fig. 2. PD distribution of safety helmet. (a) Top view; (b) 3D stereogram; (c) rotation tilt diagram

Download full size

Fig. 3. Block design. (a) Original MobileNetV3 block; (b) separate token mixer and channel mixer; (c) deep fusion block

Download full size

Fig. 4. Block diagram of ViTs-MNV3 algorithm

Download full size

Fig. 5. Curve of loss function with number of iterations

Download full size

Fig. 6. Cumulative distribution of positioning errors

Download full size

Fig. 7. Ideal environment 3D positioning result map

Download full size

Fig. 8. Point cloud triangulation. (a) Wall point cloud; (b) rotated wall point cloud; (c) wall point cloud after Delaunay triangulation

Download full size

Fig. 9. Optical power distributions and positioning errors. (a) Left wall primary reflected power; (b) right wall primary reflected power; (c) total power; (d) positioning results

Download full size

Fig. 10. Received power and positioning results. (a) Vertical simulation of optical power distribution; (b) tilt simulation optical power distribution; (c) tilted optical power matching vertical database positioning results

Download full size

Fig. 11. Experimental environment of the mine

Download full size

Fig. 12. 3D positioning results at different heights in real environment

Download full size

Fig. 13. Real environment power distribution. (a) Vertical acquisition training set power distribution; (b) skewed power distribution of the training set after vertical power transformation

Download full size

Fig. 14. Power correction effect and positioning results. (a) Comparison between the converted power and the real collected power; (b) converted power matching real acquisition power positioning result plot

Download full size

Table 1. Root mean square error and mean error for different batch sizes
View table
Table 1. Root mean square error and mean error for different batch sizes
Batch size RMSE /cm Mean error /cm Training time /min
32 12.62 11.93 11.1
64 11.66 9.01 6.8
128 10.95 8.36 4.2
256 11.42 10.15 3.7
512 12.03 12.56 2.5

Table 2. Comparison of positioning errors and complexity of different network algorithms
View table
Table 2. Comparison of positioning errors and complexity of different network algorithms
Positioning
algorithm
RMSE /
cm
Mean
error /cm
FLOPs /
10⁹
Time /
min
MobileNetV3 28.30 23.68 2.69 10.3
Transformer 25.80 22.95 1.58 4.6
ViTs-MNV3 10.95 8.36 1.42 4.2
BP 16.15 15.77 1.67 5.9
LSTM 14.80 14.61 1.46 4.5

Table 3. Channel simulation parameter

View table

Table 3. Channel simulation parameter

Symbol	Parameter	Value
$L \times W \times H$	Room size /（m×m×m）	$4.0 \times 3.0 \times 2.5$
$φ$	Tilt angle of PD /（°）	30
$ω$	Rotation angle of safety helmet /（°）	45
$γ$	Tilt angle of safety helmet /（°）	15
$A_{p}$	Physical area of photo-detector / $c m^{2}$	1
$m$	Lambert index	1
$ρ$	Reflection coefficient	0.5
$T (*)$	Gain of optical filter	1
$g (*)$	Gain of optical concertrator	2.55
$F O V$	Field of view /（°）	90
$Ф_{1 / 2}$	Semi-angle at half power /（°）	70
$P_{t}$	Average transmitted power /W	20

Table 4. Model training parameters
View table
Table 4. Model training parameters
Symbol Parameter Value
epochs Model iterations 1000
k_s Kernel size 3
e_f Expansion factor 2
o_c Output channel 128
h_s Hard-swish 1
stride Stride length 2
b_s Batch size 128

Tools

Get Citation

Copy Citation Text

Jianyong Yu, Xiaoli Hu, Qian Wang, Ling Qin, Fengying Wang, Xinchao Kou. Application of Visible Light Positioning System Based on Fingerprint Correction and Network Fusion Algorithms[J]. Acta Optica Sinica, 2025, 45(16): 1606005

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites