Fig 1.
Pipeline of LIO-CSI.
Fig 2.
Schematic diagram of label correction.
Fig 3.
Visualization of dynamic object filtering.
Fig 4.
Comparison of feature points and constraint relationships between frames: (a) edge feature points and loss le, (b) planar feature points and loss lp. Different color points represent different semantic labels.
Fig 5.
Semantic-assisted scan-context image encoding process.
Fig 6.
The sensor setup used in our experiments: (a) a Velodyne HDL-32E surround LiDAR sensor, (b) a 3DM-GX5 IMU, and (c) a Trimble BD982 GNSS receiver module with antennas.
Fig 7.
Recording platform (Volkswagen Tiguan) with sensors.
Table 1.
KITTI dataset details.
Table 2.
JLU Campus dataset details.
Table 3.
Relative pose error on KITTI dataset.
Fig 8.
Trajectory comparison on sequence 05 of the KITTI dataset.
Fig 9.
Trajectory comparison on sequence 06 of the KITTI dataset.
Fig 10.
Trajectory comparison and semantic maps of sequence 09.
Fig 11.
Vertical global consistency comparison.
Table 4.
Relative pose error of LIO-SAM and LIO-SAM-ODOM.
Table 5.
Relative pose error of LIO-SAM-ODOM and LIO-CSI.
Fig 12.
USGS national map imagery and mapping results of LIO-CSI.
The green track in USGS National Map imagery is the actual trajectory of Tiguan Volkswagen movement.
Table 6.
Relative pose error on JLU campus dataset.