A voxel-based multiview point cloud refinement method via factor graph optimization
Our proposed framework incorporates two innovative designs, namely the multi-scale voxel-grid optimization strategy and the quaternion-based factor graph optimization that employs a hybrid of point-to-plane and point-to-point factors.
- The multi-scale voxel-grid optimization strategy is mainly designed to address the problem that registration is highly non-convex and prone to getting stuck into local minima.
- The quaternion-based factor graph optimization refines the pose parameters in the tangent space by two factors, a robust constraint in structured environments provided by point-to-plane factors and a complementary constraint from non-structured surroundings provided by point-to-point factors.
The code is currently being refactoring, and will be released as soon as possible.
The code was tested on Ubuntu 18.04.
- ROS melodic
- PCL version 1.11.0 (other versions may also work)
- ceres version 2.10 or above
- Eigen 3.3x or above
Clone this repository
git clone https://github.com/WuHao-WHU/EVPA.git
cd EVPA
catkin_make
source ./devel/setup.bash
roslaunch evpa point_cloud_refinement.launch
All parameter setting can be modified in .launch
file, you can change the default setting regarding different scenarios.
You can test on the open-source TLS dataset: WHU-TLS, Robotic 3D Scan datasets, ETH Dataset.
The framework supports *.ply
, *.pcd
formats of point cloud data. You may need to transform other formats to the supported formats.
Email: haowu2021@whu.edu.cn
Do not hesitate to contact the authors if you have any question or find any bugs.