A High-Robustness 3D Visible Light Positioning Algorithm Based on Multi-Head Attention and ResNet50

Visible Light Positioning (VLP) is considered one of the most promising technologies for future Location-Based Services (LBSs). Traditional VLP systems are highly sensitive to the receiver's orientation. However, assuming that the receiver's orientation is fixed or perfectly known is unrealistic in practical applications. Thus, a random angle between the receiver and the horizontal plane inevitably appears during localization, significantly affecting positioning accuracy. To decrease position error caused by variations in the receiver's angle, this paper presents an algorithm that integrates a multi-head attention mechanism with the ResNet50 convolutional neural network. The algorithm accurately estimates the three-dimensional coordinates, comprising the receiver's X and Y coordinates and rotation angle, effectively mitigating the impact of random angular changes on positioning accuracy and significantly enhancing the system's robustness. The proposed method has been validated in a laboratory environment with dimensions of 2.6m×2.6m×2.2m and a rotation angle range of ±30°, achieving 3D positioning accuracy within 2 cm.