Research on scene matching positioning technology for large-scale indoor scenes

Abstract Indoor positioning technology is crucial for pedestrian navigation in large-scale scenes and the development of smart cities. Since Global Navigation Satellite Systems signal is usually blocked, complementary infrastructures are usually adopted to provide indoor positioning service such as pseudolites and Ultra-Wideband. However, such approaches still need additional user-side devices. Aiming at the problem, this paper presents a scene-matching based indoor positioning method for human indoor navigation, which utilizes smartphones as positioning carriers to achieve rapid and convenient positioning functionalities large-scale indoor scenes. Firstly, a prior database via the Oriented FAST and Rotated BRIEF feature extraction algorithm, employing a bag-of-words model for efficient feature matching, is proposed. Then, a Hidden Markov Model-based relocalization method uses historical information to quickly identify the actual corresponding place in similar indoor scenes, and the Perspective-n-Points are used to determinate the camera pose. Finally, experiments are designed and conducted to evaluate the performance of the approach. Test results indicate that our algorithm achieves an RMSE of 0.48 m, demonstrating improved stability and robustness in handling extreme cases.