For theory and details on implementation, Please Visit: https://medium.com/@krishnarajr319/human-pose-comparison-and-action-scoring-using-deep-learning-opencv-python-c2bdf0ddecba
Install Dependencies:
pip install -r requirements.txt
There is a test video named test.mp4 included with the repo in order to test its working.
The file called lookup_test.pickle contains the sequence of keypoints recorded for punch - side recorded from test.mp4.
In order to compare test.mp4 with the keypoints recorded in the lookup.pickle under the label punch - side,run:
python start_here.py --activity "punch - side" --video "test.mp4"
There is a file keypoints_from_video.py which can be used to create a new lookup table. In order to extract and record keypoints from test.mp4, run:
python keypoints_from_video.py --activity "punch - side" --video "test.mp4" --lookup "lookup_new.pickle"/[YOUR_LOOKUP_NAME]
Then, in order to use this new lookup, run:
python start_here.py --activity "punch - side" --video "test.mp4" --lookup "lookup_new.pickle"/[YOUR_LOOKUP_NAME]
The Posenet model used in this repo was implemented here : https://github.com/rwightman/posenet-python