A Novel Fingerprint Database Regeneration Method for Accurate Visible Light Positioning
Shiwu Xu, Fen Wei, Yi Wu
Abstract
Although the fingerprint-based visible light positioning (VLP) method can achieve centimeter-level accuracy, it requires collecting received signal strength (RSS) values from a large number of training locations in the offline stage, especially when considering the rotation angle of the photodiode (PD). To enhance the practicality of VLP based on the fingerprint method, this paper proposes a novel fingerprint database regeneration method. The proposed method does not rely on a signal propagation model but only requires knowledge of the actual coordinates of LEDs while also considering the impact from the rotation angle of the PD, making it more practical. Simulation and experimental results demonstrate that the proposed method achieves the highest positioning accuracy compared to the other four fingerprint database regeneration methods. The proposed method can still achieve satisfactory localization performance even if the proportion of training locations is low and randomly distributed. When the number of training fingerprints is 60 and randomly distributed, the proposed method can still achieve localization performance similar to that of an actual dense fingerprint database. Even when considering the rotation angle of the PD, the proposed method reduces the average positioning error by 39.85% in simulation and 29.78% in experiment, respectively, compared to sparse fingerprint databases.