lidar_cluster ROS 2 package

LIDAR pointcloud clustering in ROS 2 Humble

Build this ROS 2 package

In the following ~/ros2_ws is assumed as the ROS 2 workspace:

cd ~/ros2_ws/src

git clone https://github.com/jkk-research/lidar_cluster_ros2

cd ~/ros2_ws

sudo rosdep install --from-paths src --ignore-src -r -y

Build either with:

colcon build --packages-select lidar_cluster --symlink-install

or with optimized build:

MAKEFLAGS="-j4" colcon build --packages-select lidar_cluster --symlink-install --cmake-args -DCMAKE_BUILD_TYPE=Release

Tip

There is a detailded decription how to make this package work with ground segmentation: jkk-research.github.io/workshops/clustering_a

ROS 2 graph

graph LR;

    p[ /input_points<br/>sensor_msgs::PointCloud2]:::white --> cluster([ /cluster_node]):::light
    tf[ /tf <br/>optional topic]:::dash -.-> cluster
    cluster --> f1[ /clustered_points<br/>sensor_msgs::PointCloud2]:::white
    cluster --> f2[ /clustered_marker<br/>visualization_msgs::MarkerArray]:::white
    classDef light fill:#34aec5,stroke:#152742,stroke-width:2px,color:#152742  
    classDef dark fill:#152742,stroke:#34aec5,stroke-width:2px,color:#34aec5
    classDef white fill:#ffffff,stroke:#152742,stroke-width:2px,color:#15274
    classDef dash fill:#ffffff,stroke:#152742,stroke-width:2px,color:#15274, stroke-dasharray: 5 5
    classDef red fill:#ef4638,stroke:#152742,stroke-width:2px,color:#fff

Loading

Cluster nodes

Node	Clustering	Implementation	Description	Additional Resources
dbscan_spatial	DBSCAN	spatial	DBSCAN non-grid implementation
dbscan_grid	DBSAN	grid	DBSCAN voxel-grid-based implementation
dblane_spatial	DBlane	spatial	DBlane non-grid implementation	flowchart, notebooks
dblane_f1s	DBlane	formula	DBlane formula 1 student implementation	notebooks
euclidean_spatial	Euclidean	spatial	PCL implementation of Euclidean clustering non-grid implementation	PCL docs
euclidean_grid	Euclidean	grid	PCL implementation of Euclidean clustering voxel-grid-based implementation	PCL docs

Don't forget to source before ROS commands.

source ~/ros2_ws/install/setup.bash

ros2 run lidar_cluster dblane_spatial

ros2 launch lidar_cluster dblane_spatial.launch.py

ros2 launch lidar_cluster dblane_f1s.launch.py topic:=/input_points

Dblane-g launch with simulator

To run dblane-g with the racetrack simulator, use the Formula Student simulator package from:

https://github.com/szenergy/formula_student_packages/tree/racetracksim-extradata

Important

Switch to the racetracksim-extradata branch before building.

Example setup in ~/ros2_ws/src:

cd ~/ros2_ws/src
git clone https://github.com/szenergy/formula_student_packages.git
cd formula_student_packages
git checkout racetracksim-extradata

Build the racetrack simulator package:

cd ~/ros2_ws
source /opt/ros/humble/setup.bash
colcon build --packages-select racetrack_simulator --symlink-install
source ~/ros2_ws/install/setup.bash

Start the simulator:

ros2 launch racetrack_simulator lidar_cone_sim.launch.py

Start dblane_f1s in a new terminal (this command starts the cluster node and RViz by default):

source /opt/ros/humble/setup.bash
source ~/ros2_ws/install/setup.bash
ros2 launch lidar_cluster dblane_f1s.launch.py

Start the evaluation node in another terminal:

source /opt/ros/humble/setup.bash
source ~/ros2_ws/install/setup.bash
ros2 run lidar_cluster evaluation

Evaluation node controls:

• Press e: evaluate the currently available (already received) input data snapshot, print the metrics, and start publishing evaluation preview topics.
• Press Ctrl+C: exit the evaluator immediately.
• Published preview topics include /point_true_positives, /point_false_negatives, /point_false_positives, /edge_true_positives, /edge_false_negatives, /edge_false_positives, and /gt_edges.

Evaluator inputs (exact topics)

• /nonground_odom (sensor_msgs/msg/PointCloud2): Ground-truth cones are initialized from this stream (left/right/noise classes), so this defines what is considered the reference set.
• /interpolated_marker_map_odom (visualization_msgs/msg/MarkerArray): Predicted map points and line segments are read from this stream, so this defines what is compared against GT.

When e is pressed, the evaluator computes results only from the data that has already arrived on the two input topics at that moment.

Evaluator outputs

• Terminal output: point-level, hybrid, and edge-level metrics are printed (TP/FP/FN, precision, sensitivity, F-measure, etc.) for the current snapshot.
• /point_true_positives: GT cone points that are matched by prediction within threshold.
• /point_false_negatives: GT cone points that are currently unmatched.
• /point_false_positives: predicted points that currently do not match any GT cone.
• /edge_true_positives: GT consecutive edges that are matched by predicted edges.
• /edge_false_negatives: GT edges that are currently unmatched.
• /edge_false_positives: predicted edges that do not match GT edges.
• /gt_edges: the GT left/right edge chains in odom; useful as a stable reference overlay in RViz.

Remarks

In VS code it is advised to add the following to include path:

${workspaceFolder}/**
/opt/ros/humble/include/**
/usr/include/pcl-1.12/**
/usr/include/eigen3/**

If you are not sure where your header files are use e.g.:

find /usr/include -name point_cloud.h
find /usr/include -name crop_box.h

Images

Note

The following images shows two possible usage of clustering: an urban scenario and a race scenario. The first row shows a camera image that corresponds to the LIDAR image in the second row. In the third row an example is shown of the clustered LIDAR pointcloud with yellow.

Urban scenario	Race scenario

Citation

If you use any of this code please consider citing the paper:

@INPROCEEDINGS{10607072,
  author={Unger, Miklós and Horváth, Ernő and Pup, Dániel and Pozna, Claudiu Radu},
  booktitle={2024 IEEE International Conference on Mobility, Operations, Services and Technologies (MOST)}, 
  title={Towards Robust LIDAR Lane Clustering for Autonomous Vehicle Perception in ROS 2}, 
  year={2024},
  pages={229-234},
  keywords={Laser radar;Lane detection;Source coding;Clustering algorithms;Robustness;Pattern recognition;Autonomous vehicles;self-driving;autonomous;point cloud;LIDAR;proceeding;filter;geometric patterns},
  doi={10.1109/MOST60774.2024.00031}}