A GNSS UWB tight coupling and IMU ESKF algorithm for indoor and outdoor mixed scenario

Jie Luo,Zhengshuai Yin,Linqiu Gui
DOI: https://doi.org/10.1007/s10586-023-04208-2
2024-01-06
Cluster Computing
Abstract:Indoor and outdoor positioning systems are difficult to locate in large areas and complex environments. In order to compensate for the impact of indoor and outdoor scene transformation, an indoor and outdoor positioning system based on GNSS/UWB/IMU is proposed, which can realize seamless and high-precision positioning between buildings in mixed scenes. In view of this, GNSS and IMU fusion is used outdoors for Error-State Kalman Filter (ESKF) filtering for positioning. At the same time, this paper proposes a D-CEP algorithm to analyze the UWB ranging variance offline and improve the accuracy of UWB positioning data. In the mixed scene, the UWB and GNSS information are tightly coupled and nonlinearly optimized to improve the positioning accuracy and achieve seamless connection between indoor and outdoor staggered scenes. Indoor and outdoor scenes use UWB and IMU fusion for ESKF filtering to achieve indoor and outdoor high precision. In indoor complex scenes, due to the existence of Ultra-wideband non line of sight (UWB-NLOS) error, the observation data may be discontinuous. When the number of observed signals is less than the number required for nonlinear optimization, the UWB-NLOS error is compensated by IMU prior information, and the positioning data is obtained to make up for the lack of positioning data and realize indoor and outdoor seamless positioning. Compared with the traditional filtering method, the average positioning accuracy is improved by 40.97 % (the Mean Absolute Error is increased from 0.256336 m to 0.151308 m). Compared with the state-of-the-art LiDAR-SLAM (LIO-SAM), the approximate positioning result can be obtained (ESKF-UWB-NLOS Mean Absolute Error is 0.151308 m, LiDAR-SLAM Mean Absolute Error is 0.148853 m).
computer science, information systems, theory & methods
What problem does this paper attempt to address?