Robust GNSS/Visual/Inertial Odometry with Outlier Exclusion and Sensor’s Failure Handling

Improving the robustness of multisensor fusion state estimation is critical for applying these new techniques into practical applications. To this end, we propose outlier detection and exclusion methods for the visual data and GNSS data in a tightly coupled, sliding window optimization-based GNSS/visual/inertial odometry. To handle the complete failure of the visual data, we also propose a visual termination and keyframe fast recovery strategy. Real-world tests with multipath effect of GNSS signals and severe visual interferes are performed. Experimental results show the effectiveness of our methods in improving the robustness of the odometry, and increase of computation time due to the method are also analyzed. Attached video of the tests is available at https://www.bilibili.com/video/BV19U4y1q781?spm_id_from=333.999.0.0 .