A hackathon project initiated at MIND 2019 at Dartmouth College.
High level idea: Create useful functions that preprocess, analyze, and visualize OpenPose data. Can be used to predict and generate taxonomy of social and non-social poses and interactions.
Link to presentation goes here.
- Clone the repository.
- Install in development mode
pip install -e .
- Check out the example notebook
notebooks/Examples.ipynb - For more data, download the Sherlock OpenPose data and extract to
notebooks/output/json.
- Load data
- Plot and inspect data
- Extract just the pose_2d keypoints
- Filter based on personid
- Extract Distance Matrix across keypoints
- Normalize keypoints to common space. Scale, rotate?.
- Extract Entropy.
- Extract Synchrony.
- Extract Center of Mass per frame.
- Example of comparing distance between 2 or multiple people.
- Extract Velocity across frame.
- Extract Acceleration across frame.
https://github.com/CMU-Perceptual-Computing-Lab/openpose
https://github.com/CMU-Perceptual-Computing-Lab/openpose/blob/master/doc/output.md