Multi - LiCa 多激光雷达标定

https://github.com/TUMFTM/Multi_LiCa

Multi - LiCa

Multi - LiDAR-to-LiDAR calibration framework for ROS 2 and non-ROS applications




Introduction

This project provides an extrinsic calibration framework for quickly calibrating multiple LiDAR sensors. It employs the Generalized Iterative Closest Point (GICP) algorithm for LiDAR-to-LiDAR extrinsic calibration and uses the RANdom SAmple Consensus (RANSAC) method to calibrate the pitch and z-distance to the ground of a single LiDAR, assuming other coordinates are known.

It has proven to be robust for different sensor setups and environments without the need of an initial guess.

We use a FPFH-based feature vector creation with an TEASER++ feature matching for the coarse alignment, which is used as initial guess for the GICP algorithm.

Overview

Motion- and targetless multi - LiDAR-to-LiDAR Calibration Pipeline,
developed at the Institute of Automotive Technology, TUM

Limitations

  • Our tool was specifically developed for motionless calibration.
  • We assume that each LiDAR to be calibrated has either a directly overlapping FOV with the target LiDAR FOV or has overlap with other LiDAR(s) with overlap to the target. This can be cascading dependency to the target.
  • We assume that the ground is flat and the environment is static.
  • Input point clouds for the calibration are in sensor_msgs/PointCloud2 or in .pcd format.

Prerequisites

The bare minimum requirement for our tool is a Linux-based OS and Docker, as we provide a Docker image with our framework. You do not need to build anything locally, but you are free to do so as described in the following section. For the local build, you will need ROS 2 - humble, Python 3.10 with opend3d, scipy, ros2_numpy and pandas (optional).

Installation and Usage

🐋 Docker Environment

  1. Build the Docker image:

    复制代码
    ./docker/build_docker.sh
  2. Run the container:

    复制代码
    ./docker/run_docker.sh

🖥 Local Build

  1. Install ROS2 humble (might work with other ROS2 distributions but wasn't tested):
    https://docs.ros.org/en/humble/Installation.html

  2. Create a ROS 2 workspace:

    复制代码
    mkdir -p ~/ros2_ws
    cd ~/ros2_ws
  3. Clone the repository:

    复制代码
    git clone git@github.com:TUMFTM/Multi_LiCa.git
  4. Install dependencies:

    复制代码
    cd Multi_LiCa
    pip install --no-cache-dir --upgrade pip
    pip install --no-cache-dir -r requirements.txt
  5. Source the ROS 2 environment and build the project using colcon:

    复制代码
    source /opt/ros/$ROS_DISTRO/setup.bash
    colcon build --symlink-install --packages-up-to multi_lidar_calibrator --cmake-args -DCMAKE_BUILD_TYPE=Release

⏯️ Usage

  1. Configure the parameters to fit your data:

    复制代码
    vim config/<params-file>.yaml
  2. Launch the multi_lidar_calibrator node:

    复制代码
    ros2 launch multi_lidar_calibrator calibration.launch.py parameter_file:=/path/to/parameter/file

⚙️ Configuration

  • We provided a detailed parameter file with explanation with config/params.yaml

  • Configure config/params.yaml to fit your data. Depending on the application, you may need to specify the LiDARs, paths to .pcd files, or LiDAR topic names. You may also change GICP and RANSAC parameters.

  • In addition to LiDAR-to-LiDAR calibration, you can perform target LiDAR-to-ground/base calibration if your x,y translation and roll, yaw rotation are precisely known.

    If you are using to-base calibration, you may choose a URDF file to save the calibration so that it can be directly used in your ROS robot-state-publisher.

  • When running in a container, ensure that your local and container environments have the same ROS_DOMAIN_ID. If not, set it to be the same with export ROS_DOMAIN_ID=<ID>.

  • When using ROS 2, verify that the transformation guess is published on the /tf_static topic and that the data is published for all specified LiDARs.

🎞️ Demo

On default, the tool will launch a demo with data from OpenCalib.

It will open a window and will display three pointclouds and their initial transforms. You can inspect the files in the interactive window. After closing the window (press Q), the tool will calculate the transformations ans will print the results to the terminal, write them to the output directory and will display a windows with the transformed pointclouds.

Other OSS Calibration Frameworks

相关推荐
算家计算20 分钟前
模糊高清修复真王炸!ComfyUI-SeedVR2-Kontext(画质修复+P图)本地部署教程
人工智能·开源·aigc
虫无涯31 分钟前
LangSmith:大模型应用开发的得力助手
人工智能·langchain·llm
算家计算44 分钟前
DeepSeek-R1论文登《自然》封面!首次披露更多训练细节
人工智能·资讯·deepseek
weiwenhao1 小时前
关于 nature 编程语言
人工智能·后端·开源
神经星星1 小时前
训练成本29.4万美元,DeepSeek-R1登Nature封面,首个通过权威期刊同行评审的主流大模型获好评
人工智能
神州问学1 小时前
【AI洞察】别再只想着“让AI听你话”,人类也需要学习“适应AI”!
人工智能
DevUI团队2 小时前
🚀 MateChat V1.8.0 震撼发布!对话卡片可视化升级,对话体验全面进化~
前端·vue.js·人工智能
聚客AI2 小时前
🎉7.6倍训练加速与24倍吞吐提升:两项核心技术背后的大模型推理优化全景图
人工智能·llm·掘金·日新计划
黎燃2 小时前
当 YOLO 遇见编剧:用自然语言生成技术把“目标检测”写成“目标剧情”
人工智能
算家计算2 小时前
AI教母李飞飞团队发布最新空间智能模型!一张图生成无限3D世界,元宇宙越来越近了
人工智能·资讯