Human pose estimation is expected to use on mobile robots which need human interactions.
Cloning this repository under src folder in your ros workstation. And the same should be carried out as README.md.
$ cd $(ros-workspace)
$ cd src
$ git clone https://github.com/ildoonet/tf-pose-estimation
$ pip install -r tf-pose-estimation/requirements.txt
There are dependencies to launch demo,
- video_stream_opencv
- image_view
- ros_video_recorder : https://github.com/ildoonet/ros-video-recorder
| CMU 640x360 |
Mobilenet_Thin 432x368 |
|---|---|
 |
 |
Above tests were run on a P40 gpu. Latency between current video frames and processed frames is much lower on mobilenet version.
Source : https://www.youtube.com/watch?v=rSZnyBuc6tc
$ roslaunch tfpose_ros demo_video.launch
You can specify 'video' arguments to launch realtime video demo using your camera. See [./launch/demo_video.launch](ros launch file).