A Robust Lidar-Inertial Localization System Based on Outlier Removal

Localization and mapping are important for mobile robots. Lidar inertial system has more advantages than single lidar system to deal with rapid rotation. However, in lidar inertial systems, people often ignore the effect of outliers on track accuracy. In this paper, based on the working framework of LIO-SAM, we take a solution to deal with this problem. Firstly, we extract non-ground points from depth images. Then we cluster non-ground points and filter the clustered objects that are less than a given threshold. The features extracted from the depth image are used for scan-to-map matching to obtain the lidar odometry factor. Finally, the factor graph is used to optimize fusion. The results show that our proposed method effectively reduced the error of the estimated trajectory compared with LIO-SAM, and its accuracy is higher than those single lidar systems of LOAM and LEGO-LOAM.