Visible light positioning using image-based localization

• Main Author: Huo, Yuchong
• Other Authors: Arokiaswami Alphones
• Format: Thesis-Master by Coursework
• Language: English
• Published: Nanyang Technological University 2023
• Subjects: Engineering::Electrical and electronic engineering::Optics, optoelectronics, photonics
• Online Access: https://hdl.handle.net/10356/169550
• Institution: Nanyang Technological University

With the widespread application of visible light communication, low-cost and efficient visible light positioning technology has gradually become the focus of indoor location. However, existing methods in this field still have certain limitations. In this paper, a novel high-precision indoor VLP algorithm is proposed, which combines the traditional VLP method and the computer vision technology. It succeeds in realizing receiver’s position and pose estimation based on the image of a single LED light source. In particular, it first obtains the LED’s information in the world coordinate system by analyzing the time-domain information transmitted by VLC. After that the spatial-domain information of the lamp can be acquired by extracting features from the pictures with image processing technology, which also performs well in pre-processing for the image to further improve the feasibility of this algorithm. Then, based on the perspective projection principle and solid geometry theory, the coordinate of the LED in the camera coordinate system is calculated. Combining the above two information, the final step is to use the linear least square to estimate the position and pose of the camera. The proposed algorithm can break the reliance of traditional VLP methods on multiple light sources. Meanwhile,due to the use of VLC, it can function as well without the requirement to know the corresponding relationship of projections in advance, which is significant for some computer-vision methods like PnL. The experimental results show that when the camera set at heights of 96 cm and 133 cm, the average positioning error is 5.99 cm and 4.59 cm, and the maximum error is controlled within the range of 10 cm. At the same time, it has greater orientation angle estimate accuracy when compared to current V-P4L methods, which shows great accuracy of the proposed method.