OpenPose ROS

This work is develped from here. I modified most of files to make it support the latest (commit number 5295f2f) OpenPose library C++ API as well as optimized the code to speed up the wrapper to arrive 10 fps on GTX 1060.

When a depth image is synchronized with the RGB image (RGB-D image), a 3d extractor node has been implemented to obtain 3d pose estimation from the 2d pose estimation gien by OpenPose through the projection of the 2d pose estimation onto the point-cloud of the depth image. Also, a visualization node for the 3d results has been implemented.

NOTE: Instead of doing painful following installation steps, I provide docker image file which support OpenPose development environment as well as OpenGL to do visualization, you can directly access it from here. Or you can just run the following command to get the docker image:

docker pull bluebirdpp/openpose:latest

Install OpenPose Library

Install openpose and its dependencies at here.

Potential issues about OpenPose Installation

• Your OpenPose would better build with the same version of OpenCV which comes with ROS, and ROS built-in OpenCV might located in /opt/ros/kinetic/include. Otherwise it would cause some problems when operating with image.

• If you choose to build OpenPose by your own custom caffe, you should be careful with your caffe's version. The latest OpenPose Library (commit number 5295f2f) is only compatible with caffe before Aug 14, 2018 (commit number 4353628).

Running OpenPose ROS Wrapper and 3d Pose Extractor

NOTE: I used ASUS Xtion Pro Live camera, and run it by openni2_launch default ROS parameter, so the image resolution is 480*640. If you get images stream in different resolution, you should modify the following two lines in skeleton_extract_3d_node.cpp manually to match you image resolution.

Now you can run the code. First connect a RGB camera and run the corresponding ROS drivers to start to publish the images (they must be image_raw). For example I used ASUS Xtion Pro Live camera and my command was:

roslaunch openni2_launch openni2.launch

You should get RGB and depth streams in rqt_image_view.

Then, initialize the 2d detection service node:

rosrun openpose_ros_pkg openpose_ros_node_3d

If everything works fine you should see the following output in your shell:

[INFO] [1501140533.950685432]: Initialization Successful!

Finally, to get the 2d poses of the images from the camera and visualize the output, run:

rosrun skeleton_extract_3d skeleton_extract_3d_node_test

Then, images stream with markered skeleton keypoints should be published on topic /openpose_ros/skeleton_3d/detected_poses_image, you can visualize it in rqt_image_view, and 3D skeleton keypoints information should published on topic /openpose_ros/skeleton_3d/detected_poses_keypoints_3d.

Visualize 3D skeleton keypoints on Rviz

If you want to change the color, shape, size etc. of the markers, you need to go to /skeleton_extract_3d/src/skeleton_extract_3d_visualization_node.cpp to modify related code manually. To see the options you have go here. To set the options of the bodyjoints:

To set the options of the skeletons:

Then you can run the visualization node:

rosrun skeleton_extract_3d skeleton_extract_3d_visualization_node

Finally, start Rviz to visualize 3D skeleton tracking results (set /camera_depth_frame as world frame):

rosrun rviz rviz -f /camera_depth_frame