install

• Install Python depend packages

  pip install -r requirements.txt

• Donwload model data

  cd ./lightweight_openpose_ros2/datas/

  wget https://download.01.org/opencv/openvino_training_extensions/models/human_pose_estimation/checkpoint_iter_370000.pth

• Install depend package

  rosdep install -y -i --from-path src

• Build package

  colcon build --symlink-install --packages-up-to lightweight_openpose_ros2

Execute

2D Pose Estimation (lightweight_openpose_ros2)

ros2 run lightweight_openpose_ros2 lightweight_openpose_ros2 --ros-args -r /image_raw:=<RGB IMAGE TOPIC NAME>

• Parameters

  params name	type	default	description
  checkpoint_path	string	.../checkpoint_iter_370000.pth	Path to the model checkpoint file.
  device	string	"cpu"	Computation device ("cpu" or "cuda").
  height_size	int	256	Input image height for the network.
  qos.reliability	string	"RELIABLE"	QoS reliability policy ("RELIABLE" or "BEST_EFFORT").
  qos.durability	string	"VOLATILE"	QoS durability policy ("VOLATILE" or "TRANSIENT_LOCAL").
  qos.depth	int	10	QoS history depth.
  debug	bool	true	Whether to show the debug window with detection results.

• Service
  This node does not start detection automatically. You need to call the execute service to start/stop detection.

  • Start Detection

    ros2 service call /execute std_srvs/srv/SetBool "{data: true}"

  • Stop Detection

    ros2 service call /execute std_srvs/srv/SetBool "{data: false}"

3D Pose Estimation (lor_transformer)

This package converts 2D poses to 3D using depth images and camera info. Run the node with remappings to your camera topics.

ros2 run lor_transformer lor_transformer --ros-args \
    -r image_raw:=/camera/aligned_depth_to_color/image_raw \
    -r depth_camera_info:=/camera/aligned_depth_to_color/camera_info \
    -r color_camera_info:=/camera/color/camera_info \
    -p target_frame:=base_link

![NOTE] Replace /camera/... with your actual RealSense/Depth camera topics.*

• Parameters

  params name	type	default	description
  depth_scale	double	0.001	Scale factor to convert depth image values to meters (e.g., 0.001 for mm to m).
  depth_sample_size	int	5	Window size (NxN) for median depth filtering around the keypoint.
  target_frame	string	""	Target TF frame for 3D poses. If empty, uses the depth optical frame.

• Visualization The node publishes MarkerArray to human_pose_markers which can be viewed in RViz2. The node publishes MarkerArray to human_pose_markers which can be viewed in RViz2.

Demo

ros2 launch lightweight_openpose_ros2 demo_launch.py