Skip to content

url-kaist/Quatro-LeGO-LOAM

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

82 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Quatro-LeGO-LOAM

Robust Global Registration Quatro (22' ICRA) + LeGO-LOAM (18' IROS)

  • For more details for each algorithm,
    Quatro https://github.com/url-kaist/quatro
    LeGO LOAM https://github.com/RobustFieldAutonomyLab/LeGO-LOAM
  • In Quatro registration example, we used fast point feature histogram (FPFH), Patchwork, and etc. to reduce the computational time of feature extraction & matching, i.e. the front-end of global registration, from tens of seconds to almost 0.2 sec.
  • But note that Quatro-LeGO-LOAM only uses the FPFH descriptor to perform feature matching and then estimates a relative pose through Quatro.
  • LeGO-LOAM has a large drift, which is accumulated in large-scale maps. For this reason, loop detection using Radius-Search may not generate the loop closure constraint properly. Therefore, if the drift is larger than the used KITTI 05 sequence, it is better to use another descriptor or another Odometry method. Then, You can use Quatro-SC-LeGO-LOAM or Quatro-Faster-LIO alterlatively. It hasn't been released yet.😭

Test Env.

The code is tested successfully at

Build Quatro

mkdir -p ~/catkin_ws/src
cd ~/catkin_ws/src
git clone [email protected]:url-kaist/quatro.git
cd quatro && mkdir build && cd build
# To build Quatro, `pmc-src` should be placed in `build` directory in catkin workspace
# i.e. `~/catkin_ws/build/pmc-src`
cmake ..
mv pmc-src/ ../../../build/
cd ~/catkin_ws
catkin build quatro 

Note that without pmc-src, the below error occurs!

CMake Error at quatro/CMakeLists.txt:53 (add_subdirectory):
  add_subdirectory given source "~/catkin_ws/build/pmc-src" which
  is not an existing directory.

How to run

cd ~/catkin_ws/src
git clone {this repo}
cd ..
catkin_make
source devel/setup.bash
roslaunch lego_loam run.launch

Dataset

KITTI 05 sequence dataset

  • Download KITTI 05 sequence dataset with following command. We already make a rosbag file as an example (15.6GB)
wget https://urserver.kaist.ac.kr/publicdata/quatro/kitti_sequence_05.bag
  • You can run the KITTI bag file as follows:
rosbag play kitti_sequence_05.bag --clock --topics /kitti/velo/pointcloud

Else

  • To generate more bags using other KITTI raw data, you can use the python script kitti2bag.
  • Check the parameters according to the various lidars in utility.h : VLP-16 HDL-32E HDL-64E VLS-128 OS1-16 OS1-64 OS0-128
  • You can run your bag file as below code.
rosbag play {your_bag}.bag --topics {pointcloud_topic_message}

Trajectory Evaluation

  • I used evo (link) to evaluate the trajectories.
  • You can install evo using the following command
pip install evo --upgrade --no-binary evo
  • First, you have to modify the directory traDirectory in utility.h file.
// at 62 line of utility.h
extern const string traDirectory = "/home/{your_dir}/lego_loam_trajectory/";
  • The following command can be used to evaluate the trajectory.
evo_traj kitti {your_file_name}.txt --ref={kitti_groundtruth_file_name}.txt -p --plot_mode=xyz
  • Also, You can use other methods to evaluate your trajectories e.g. metric of trajectory.
mkdir results
evo_ape kitti [kitti_file_name].txt [your_file_name].txt -va --plot --plot_mode xyz --save_results results/[name].zip
evo_ape kitti [kitti_file_name].txt [your_file_name].txt -va --plot --plot_mode xyz --save_results results/[name]
  • You can also save final PCD file. Modify your pcdDirectory in utility.h file.
// at 60 line of utility.h
extern cost string pcdDirectory = "/home/{your_dir}"

Example

  • Compare LeGO-LOAM and Quatro-LeGO-LOAM with KITTI 05 dataset. We used cloudcompare to visualize pcl.
  • You can download cloudcompare with folloing command.
snap install cloudcompare

Compare PointCloud

  • Compare by CloudCompare. PointCloud

  • From the gif below, red : source, green : target, blue : estimation. video

  • Even though they are the same Radius Search and ICP parameters, they differ greatly depending on Quatro's Initial Guess.

  • Existing Lego-loam does not satisfy the icp threshold due to the distance between src and tgt, but using ICP after Quatro produces much better ICP results.

Compare Trajectory with GroundTruth

  • Quatro-Lego-Loam's Trajectory evaluation showed an error of less than 10cm !!
Dataset scene1 scene2
KITTI 05 seq. scene1 scene2

Visualize Quatro Process

  • You can visualize Quatro's source, target, estimation clouds : /quatro_src, /quatro_tgt, /quatro_est.
  • Below visualization picture is an example. red : source, green : target, blue : estimation

About

No description, website, or topics provided.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published