Cuboid_detector

ROS 2 repository for 6D pose estimation of cuboid-like objects

1. 1. Setup
   1. Installation from Docker
   2. Installation from Source
2. 2. Commands for Demo
   1. Evaluation Mode
   2. Online Mode (ROS2 Topic)
3. 3. (Optional) How to save ros2 bag to data files

1. Setup

Installation from Docker

docker build -t cuboid_detector:latest .

Installation from Source

• The following setting is tested for Intel® RealSense™ LiDAR Camera L515 with ROS 2 Humble in ubuntu 22.04 only.
• Build the projects:

# librealsense를 먼저 설치

# https://github.com/IntelRealSense/realsense-ros?tab=readme-ov-file#installation-on-ubuntu

cd YOUR_ROS_WORKSPACE/src # Cuboid_detector 폴더 위치시키기

ros2_ws (YOUR_ROS_WORKSPACE)
├── build
├── install
├── log
└── src
    ├── Cuboid_detector
        ├── cuboid_detector
        │   ├── dataset
        │   │   ├── camera_info
        │   │   ├── data
        │   │   │   └── color_000000.png
        │   │   │   └── ...
        │   │   │   └── color_000009.png
        │   │   │   └── pose.json
        │   │   └── mesh
        │   │       └── danpla_box.obj
        │   │   └── ...
        │   ├── script
        │   │   ├── cuboid_detector.py      # Main detection script
        │   │   ├── gt_annotator.py         # Data annotation
        │   │   └── ...
        │   └── ...
        ├── README.md
        └── requirements.txt

sudo apt install ros-humble-vision-msgs ros-humble-realsense2-camera-msgs
cd Cuboid-detector && pip3 install -r requirements.txt
cd ../.. && colcon build
source install/local_setup.bash

2. Commands for Demo

Evaluation Mode

(Offline test with saved PNG images)

Basic:

cd YOUR_ROS_WORKSPACE/src/Cuboid_detector/cuboid_detector/script

python3 cuboid_detector.py --evaluate --visualize

Docker:

# New container
xhost +
docker run -ti --net host -v /dev/shm:/dev/shm \
  -e DISPLAY=$DISPLAY --privileged -v /tmp/.X11-unix:/tmp/.X11-unix \
  cuboid_detector:latest \
  bash

# In the container
cd src/cuboid_detector/script
cuboid_detector cuboid_detector --evaluate --visualize

The evaluation results will show up like this:

• Green (thickness=2): Ground truth
• Red (thickness=4): Predicted bounding box

# Error in box frame
Mean x error: 1.8691175989213542 mm
Mean y error: -0.6178012588862275 mm
Mean z error: -5.399375062604473 mm
Mean translation error: 6.131211014944094 mm
Mean rotation error: 0.6364534439764666 degrees

Online Mode (ROS2 Topic)

(Real-time operation with L515 camera)

Basic:

# L515 카메라 실행 또는 ros2 bag play YOUR_ROS_BAG_FILE_PATH
ros2 run cuboid_detector cuboid_detector

Docker:

docker run -ti --net host -v /dev/shm:/dev/shm cuboid_detector:latest

Check the node status:

$ ros2 node info /cuboid_detector
/cuboid_detector
  Subscribers:
    /L515/color/camera_info: sensor_msgs/msg/CameraInfo
    /L515/color/image_raw: sensor_msgs/msg/Image
    /L515/depth/camera_info: sensor_msgs/msg/CameraInfo
    /L515/depth/image_rect_raw: sensor_msgs/msg/Image
    /L515/extrinsics/depth_to_color: realsense2_camera_msgs/msg/Extrinsics
  Publishers:
    /cuboid_detections: vision_msgs/msg/Detection3DArray
    /parameter_events: rcl_interfaces/msg/ParameterEvent
    /rosout: rcl_interfaces/msg/Log
  Service Servers:
    /cuboid_detector/describe_parameters: rcl_interfaces/srv/DescribeParameters
    /cuboid_detector/get_parameter_types: rcl_interfaces/srv/GetParameterTypes
    /cuboid_detector/get_parameters: rcl_interfaces/srv/GetParameters
    /cuboid_detector/list_parameters: rcl_interfaces/srv/ListParameters
    /cuboid_detector/set_parameters: rcl_interfaces/srv/SetParameters
    /cuboid_detector/set_parameters_atomically: rcl_interfaces/srv/SetParametersAtomically
  Service Clients:

  Action Servers:

  Action Clients:

3. (Optional) How to save ros2 bag to data files

cd YOUR_ROS_WORKSPACE/src/Cuboid_detector/cuboid_detector/script

python3 convert_topic_to_file.py
ros2 bag play YOUR_ROS_BAG_FILE_PATH