This repository implements C3P-VoxelMap, a compact, cumulative, and coalescible probabilistic voxel mapping method to enhance performance, accuracy, and memory efficiency in LiDAR odometry. Based on VoxelMap, our work reduces memory consumption by two strategies:
- Compact point-free representation for probabilistic voxels and cumulative update of the planar uncertainty without caching original point clouds.
- On-demand voxel merging by taking advantage of the geometric features in the real world, accumulating voxels in a locality-sensitive hash and triggers merging lazily.
The paper is available on arxiv: C3P-VoxelMap: Compact, Cumulative and Coalescible Probabilistic Voxel Mapping
ROS (tested on Noetic)
PCL (>= 1.8)
Eigen (>= 3.3.4)
livox_ros_driver, follow livox_ros_driver Installation.
Clone the repository and compile it by catkin_make:
mkdir -P ~/catkin_ws/src & cd ~/catkin_ws/src
git clone https://github.com/deptrum/c3p-voxelmap.git
cd ..
catkin_make
source devel/setup.bash
To run on dataset (KITTI dataset for example), firstly edit configuration file config/velodyne.yaml.
Besides general parameters such as point cloud topic name, there are some extra configutations concerning on-demand voxel merging. To enable voxel merging, set voxel_merging/enable_voxel_merging to true. The other parameters in voxel_merging can also be configured to adjust the effectiveness of on-demand voxel merging.
Optionally, set visualization/pub_merged_voxel to true to visualize voxel merging results. Merged voxel planes are shown with the same color in visualization.
After setting parameters, run the ROS package:
cd ~/catkin_ws
source devel/setup.bash
roslaunch c3p_voxelmap mapping_velodyne.launch
In the meanwhile, play rosbag and the visualization results will be shown in RViz window.
Thanks for VoxelMap.