Keypoint extractor using openpose and opencv.

Given a video generates a numpy array containing the extracted keypoints and cleans up afterweards, if wished.

Navigate to this repository. This will install the tool vid_to_key and link it to $HOME/.local/bin:

pip install -e .

Add -vvv flag to see all installation outputs.

If you cannot find the executable vid_to_key afterwards and/or adding -vvv does not produce an output line linking script, try the following:

python setup.py install

Now it can be called from anywhere on you system. If you installed it in a virtualenv, make sure to start the virtualenv before calling the script.

If run on hci servers needs to be run on hcigpu01!

This will extract keypoints from all videos or images inside the folder using the gpu with index 1:

cd /path/to/where_my_video_is/
vid_to_key -cvd 1

It is equivalent to:

vid_to_key --v '/path/to/where_my_video_is/'

If fthe folder contains videos, keypoints are extracted for each video. If it contains images, these are interpreted as coming from one video. Make sure the images follow the nameing convention FRAMENUMBER_VIDEONAME.jpg, though only the framenumber and the _ are important.

Note that if the folder contains both videos and images an error is raised.

You can also point to a video directly:

vid_to_key --v /path/to/video.mpg

The script assumes that there is only one person in the video. This number can be increased with the --np argument:

vid_to_key --np 3 --v /path/to/video.mpg

During the creation of the keypoint file a lot of images ans json files are created. These can be deleted using the --ci and --cj flags respectiveley.

vid_to_key --ci --cj --v /path/to/video.mpg