Skip to content

ROS 2 LIDAR Clustering

The goal of this short practice/workshop is to demonstrate the filtering of LIDAR data into objects. Thus, we create larger objects/clusters from individual LIDAR points. These objects can be pedestrians, cars, buildings, etc. The practice is compatible with ROS 2. Static Badge

Requirements (high-level overview)

  1. ROS 2 Humble: 🟠 see previous materials or docs.ros.org/en/humble/Installation.html
  2. Log file with raw LIDAR data (MCAP format, bag) ✅
  3. The patchworkpp package for ground plane filtering ✅
  4. The lidar_cluster package for clustering execution ✅

Video Overview

The following screen recording demonstrates the necessary steps:

Step 1. - Download the raw data

To cluster LIDAR data, we first need LIDAR data. Use one of the following 3 options.

Download MCAP [~540MB]

In our examples, the .mcap file is saved to the /mnt/c/bag/ folder. If you want to use another directory, please modify accordingly.

Option B: Download MCAP via your terminal

Don't forget to change the directory first. In our case, `/mnt/c/bag/` is the place where we will put it: 
cd /mnt/c/bag/

wget https://laesze-my.sharepoint.com/:u:/g/personal/herno_o365_sze_hu/Eclwzn42FS9GunGay5LPq-EBA6U1dZseBFNDrr6P0MwB2w?download=1  -O lexus3-2024-04-05-gyor.mcap

Option C: Use your own MCAP

You can use your own MCAP file, but in this case, you need to modify the following:

  • The LIDAR topic
  • In our example, this is /lexus3/os_center/points
  • LIDAR frame
  • In our example, this is lexus3/os_center_a_laser_data_frame

Later, don't forget to update these in the subsequent steps.

Verify the raw data

Play the bag with a command similar to the following: 

ros2 bag play /mnt/c/bag/lexus3-2024-04-05-gyor.mcap -l

Info

The -l switch in the play command means loop playback.

Success

If everything works as expected, you should see multiple topics in another terminal

Topics In another terminal, issue the following command:

ros2 topic list
You should see a similar topic list:

/clock
/events/read_split
/lexus3/gps/duro/current_pose
/lexus3/gps/duro/imu
/lexus3/gps/duro/mag
/lexus3/gps/duro/navsatfix
/lexus3/gps/duro/status_flag
/lexus3/gps/duro/status_string
/lexus3/gps/duro/time_diff
/lexus3/gps/duro/time_ref
/lexus3/os_center/points
/lexus3/os_left/points
/lexus3/os_right/points
/lexus3/zed2i/zed_node/left/image_rect_color/compressed
/parameter_events
/rosout
/tf
/tf_static

Also, there must be at least one sensor_msgs/msg/PointCloud2, check with: 

 ros2 topic type /lexus3/os_center/points
Result: 
sensor_msgs/msg/PointCloud2

Step 2 - Install ROS 2 packages

Info

If you don't have a ~/ros2_ws/ workspace, you will need the following command: 

mkdir -p ~/ros2_ws/src
If you have a different workspace name, modify the following commands accordingly.

Clone patchworkpp package

The patchwork-plusplus-ros is the ROS 2 package of Patchwork++ (@ IROS'22), providing fast and robust LIDAR ground segmentation. We recommend the JKK-research fork, which includes some improvements, or you can use the original KAIST version.

cd ~/ros2_ws/src

git clone https://github.com/jkk-research/patchwork-plusplus-ros
or 
git clone https://github.com/url-kaist/patchwork-plusplus-ros -b ROS2

Clone lidar_cluster package

cd ~/ros2_ws/src

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

Build the packages

cd ~/ros2_ws

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

Step 3 - Execution

What is the expected operation?

graph TD;

    p1[ /lexus3/os_center/points<br/>sensor_msgs::PointCloud2]:::white --> patchwork([ /patchwork_node]):::light
    patchwork --> p
    p[ /nonground<br/>sensor_msgs::PointCloud2]:::white --> cluster([ /cluster_node]):::light
    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
Don't forget to source 
source ~/ros2_ws/install/setup.bash

ros2 bag play /mnt/c/bag/lexus3-2024-04-05-gyor.mcap -l


ros2 launch patchworkpp demo.launch.py  cloud_topic:=/lexus3/os_center/points cloud_frame:=lexus3/os_center_a_laser_data_frame
Use one of the following clustering algorithms:

ros2 launch lidar_cluster dbscan_spatial.launch.py
DBSCAN (Density-Based Spatial Clustering of Applications with Noise) is a non-grid-based clustering algorithm. On a modern 6-core or better CPU, you can expect at least 10 Hz performance.

ros2 launch lidar_cluster euclidean_spatial.launch.py
Non-grid clustering based on Euclidean distance. On a modern 6-core or better CPU, you can expect at least 5 Hz performance. 

ros2 launch lidar_cluster euclidean_grid.launch.py
Voxel grid-based clustering based on Euclidean distance. On a modern 6-core or better CPU, you can expect at least 100 Hz performance.

ros2 launch lidar_cluster rviz02.launch.py

Success

If everything works as expected, you should see a similar rviz window. lidar_cluster01