A Filter-Based Integration of GNSS, INS, and Stereo Vision in Tight Mode With Optimal Smoothing

In recent years, it is increasingly prevalent to integrate global navigation satellite system (GNSS) with visual-inertial navigation systems (VINSs) to provide high-precision and continuous navigation solutions. However, its accuracy and robustness are challenged by GNSS signal blockages, inertial navigation system (INS) error accumulation, and noisy visual measurements, especially in severe environments, such as urban downtowns. To achieve accurate and consistent pose estimation in complex scenarios, we innovatively propose a filter-based GNSS/INS/vision integrated system, which fuses GNSS pseudorange, carrier phase, inertial measurement unit (IMU), and visual measurements at the observation level. Meanwhile, the full set of measurements collected by heterogeneous sensors is utilized for optimal pose estimation through the Rauch–Tung–Streibel smoother (RTSS)–forward–backward smoother (FBS) smoother, which can simultaneously enhance the navigation performance in terms of both relative and absolute accuracy in the postprocessing mode. The results of vehicular experiments in both GNSS partly blocked and challenging environments indicate that the proposed system can achieve decimeter-level to centimeter-level position accuracy without obvious convergence processes. Combining the advantages of the tight integration mode and effective smoother, the proposed method could improve the position accuracy by about 80% [root mean square (rms) error reduction] compared to the state-of-the-art VINS-Fusion. Meanwhile, noticeable improvements can also be obtained in attitude estimation.