Republish RGBD Data from Realsense to LCM

• 本项目通过 python 代码直接调用 realsense 深度相机，获取 RGB 帧和 Depth帧

• 将 RGBD 数据打包为 LabelFusion 所需要格式的 lcm 消息并发布到 OPENNI_FRAME 通道

• 使用 lcm-logger 收集原始数据，用于 LabelFusion 6D物体位姿标注

Environmen Setup

一、Realsense 深度相机配置

1. 安装 librealsense 从而使用 realsense-viewer 检测相机

   sudo mkdir -p /etc/apt/keyrings
   curl -sSf https://librealsense.intel.com/Debian/librealsense.pgp | sudo tee /etc/apt/keyrings/librealsense.pgp > /dev/null
   
   echo "deb [signed-by=/etc/apt/keyrings/librealsense.pgp] https://librealsense.intel.com/Debian/apt-repo `lsb_release -cs` main" | \
   sudo tee /etc/apt/sources.list.d/librealsense.list
   sudo apt-get update
   
   sudo apt-get install librealsense2-dkms
   sudo apt-get install librealsense2-utils
   sudo apt-get install librealsense2-dev
   sudo apt-get install librealsense2-dbg

   打开终端，运行 realsense-viewer 命令检测相机。

2. Python Packages

   pip install pyrealsense2

二、安装 LCM (via pip)

LCM can be installed via the Python package manager (pip) on many systems. To do so, run:

pip3 install lcm

This package contains:

• The LCM Python module

• LCM executables (for example, lcm-logplayer)

  Note: Java-based executables (like lcm-logplayer-gui) are not included for musl-based linux distributions

• Development files (headers and libraries)

Note: this package has a hard runtime dependency on GLib 2.0. If you have not already, please install this dependency before using the Python package.

三、安装 Python Packages

pip install numpy opencv-python

收集原始数据

1. 运行 scripts/lcm_republisher.py，启动相机发布 lcm 消息

2. 启动终端运行以下命令：

   cd /path/to/data-folder
   lcm-logger

   Ctrl + C 结束记录日志

物体 6D 位姿标注

1. Install Nvidia Toolkit: install-guide

2. git clone https://github.com/Y-pandaman/labelfusion-cuda11.3.git

3. LabelFusion/docker/docker_run.sh /path/to/data-folder

   The docker_run.sh script calls nvidia-docker to start the LabelFusion Docker container with an interactive bash session. The first time it runs the LabelFusion image will be downloaded from DockerHub automatically.

   命令会将本地的 LabelFusion 源目录和数据目录（如果提供了路径）挂载为 Docker 内的卷。Docker 容器内的路径为：

   ~/labelfusion <-- the mounted LabelFusion directory

   ~/labelfusion/data <-- the mounted data directory

4. 数据标注流程参考 Quick Pipeline Instructions

   请仔细阅读 Data Organization 参考文档，其中提供了 Object meshes 的存放位置及所需信息 object_data.yaml 的组织方式。

   请注意 lcmlog 的存放位置：

   ~/labelfusion/data/logs/
       logs/
           2025-01-01-01/
               original_log.lcmlog
   

   缺少子文件夹可能导致运行命令 run_alignment_tool 时找不到所需文件。

---

相关链接

https://github.com/RobotLocomotion/LabelFusion

https://github.com/lcm-proj/lcm

https://github.com/MobileManipulation/rgbd_ros_to_lcm