Dual LCC-Net: Targetless LiDAR-Camera Calibration with Multiple LiDARs

Preopocessing

Description

This repository contains code for our APPRAS project conducted in the WS23 by:

• Bryan Oppong-Boateng
• Daniia Vergazova
• Mehmet Yasin Cifci
• Sönke Nickelsen
• Ziad Abouhalawa

We have implemented an extension of the LCC-Net [1] architecture for usage on a custom real world dataset with two lidar inputs and evaluated it's performance. The results can be seen in the technical report.

Installation

1. Install ROS Humble.
2. Install rosdep
3. Install colcon
4. Install the Git LFS client appropriate for your operating system.
   • If you already cloned the repository and you want to get the latest LFS object that are on the remote repository, such as for a branch from origin:

   git lfs fetch origin main
   

5. Clone this repository.
6. Move into workspace directory.

   cd workspace

7. Install ROS dependencies.

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

8. Build package.

   colcon build

9. Source the workspace.

   source install/setup.sh

Process Launching Guide

Two primary workflows are defined to cover the entire process flow from data synchronization to processing and evaluation. These workflows are intended for typical use and combine multiple steps into a single launch file:

Main Workflows:

For the inference process: Launches the workflow from synchronizing and processing data to performing inference.

ros2 launch automatic_targetless_camera_lidar_calibration sync_process_then_infer.launch.py

For the export process: Initiates the workflow from synchronizing and processing data to exporting it in the KITTI format.

ros2 launch automatic_targetless_camera_lidar_calibration sync_process_then_export.launch.py

Starting Individual Nodes:

In addition to the main workflows, individual nodes can be launched for specific tasks.

To synchronize and process sensor data:

ros2 launch automatic_targetless_camera_lidar_calibration sync_and_process.launch.py

To export processed data to the KITTI format:

ros2 launch automatic_targetless_camera_lidar_calibration export_to_kitti_format.launch.py

To perform inference on processed data:

ros2 launch automatic_targetless_camera_lidar_calibration inference.launch.py

Model

Once the data is preprocessed, the model can be trained following usign the properLCCNet repo following this guide.