基 于 到 达 角 的 室 内 可 见 光 指 纹 自 修 正 定 位 方 法

Objective Indoor visible light positioning technology has potential advantages in advanced applications such as position-assisted channel estimation, drone swarms, and fast-moving robots, which requires centimeter-level positioning accuracy and positioning frequencies of hundreds of hertz. In real-world scenarios where the receiver is tilted, the uncertainty in the receiving direction leads to random variations in received signal strength (RSS), which significantly reduces the accuracy of the vertical RSS-based fingerprint positioning method and may even cause the positioning function to fail. When the receiver is tilted, in the offline stage, constructing the RSS fingerprint database that represents the position of the fingerprint point and different tilt angles of the receiver requires repeatedly collecting a large number of RSS values at various tilt angles for each fingerprint point. The labor and time costs of this traditional construction method become difficult to estimate as the positioning area, the number of positioning light sources, and the granularity of the fingerprint database increase. This greatly reduces the practicality of fingerprint positioning. To improve the accuracy and practicality of fingerprint positioning in actual tilted scenarios, we propose a fingerprint self-modified positioning method based on the angle of arrival (AOA). Methods The research methods for indoor RSS-based fingerprint positioning in this paper are primarily based on the vertical receiving path loss (PL) exponent model. A variable gain factor is introduced to modify the vertical receiving PL exponent model. A fine-grained vertical fingerprint database is generated by measuring the vertical RSS of a few fingerprint points, and the vertical fingerprint database is self-modified according to the AOA. The tilted fingerprint database, with the same dimensions and granularity, is obtained, and the weighted k nearest-neighbor matching algorithm is used to achieve high precision and fast positioning. Results and Discussions The experimental results show that the RSS obtained by the modified model of the tilted receiving PL exponent is highly consistent with the actual RSS. Additionally, the tilted fingerprint database obtained through the modified method proposed in this paper can replace the actual tilted fingerprint database for position matching (Fig. 7). Based on the vertical RSS measured and the tilted RSS from 265 locations, the vertical and tilted fingerprint databases with grid sizes of 10 cm×10 cm and 20 cm×20 cm are constructed. The polar angle θ of the tilted fingerprint database is 30.24° and the azimuth angle ω is 321.06° . With a grid size of 10 cm×10 cm, the cumulative distribution comparison results of positioning errors based on actual RSS fingerprints, modified RSS fingerprints, and unmodified RSS fingerprints show that the positioning accuracy based on self-modified RSS fingerprints is very close to that based on actual RSS fingerprints, and the vertical fingerprint database is not suitable for positioning when the receiver is tilted (Fig. 11). Using 4 LEDs, when the grid sizes of the fingerprint database are 20 cm×20 cm and 10 cm×10 cm, the root mean square errors (RMSEs) of positioning are almost unchanged when the yaw angle α changes under the same polar angle θ, and the RMSEs are almost unchanged with increasing polar angle θ when the polar angle is less than 31°, and the positioning RMSEs are about 3.04 cm and 1.57 cm. When the polar angle θ exceeds 35°, RMSEs increase dramatically, and it is difficult to improve positioning accuracy by reducing the grid size of the fingerprint database (Fig. 12). When the number of LEDs increases to 8, the azimuth angle is arbitrary and the polar angle is 56°, the RMSEs drop to 3.96 cm and 1.97 cm, which are close to the RMSEs reached when the polar angle is less than 31° in Fig. 12 (Table 3). Under 5 LEDs, the average error is comparable to the improved SV2 algorithm based on the actual tilted fingerprint database and better than the CNN algorithm. When the number of LEDs reaches 6 with a symmetrical distribution, the average error is about 1.77 cm, which is better than the improved SV2 algorithm, and the maximum positioning error is about 10.49 cm, which is comparable to the improved SV2 algorithm. Conclusions In this paper, we propose a self-modified fingerprint positioning method based on AOA. This method can quickly generate a tilted fingerprint database with the same dimensions and granularity as the vertical fingerprint database, based on the AOA of the receiver and the current receiving direction. In addition, it avoids the need for repeated collection of multi-dimensional feature fingerprints, including RSS and AOA, during the offline stage. The time complexity of the online positioning matching process is reduced. For large-scale real-world scenarios with numerous LEDs, the proposed positioning method can maintain stable centimeter-level positioning over a wide range of AOAs, which dramatically improves the practicability and accuracy of the positioning system.