Use mediapipe holistic model to classify pose application.
This repository use mediapipe to get the body pose of 33 key points (coordinate points) which create a dataset to be classified the body pose.
The idea is from AI Body Language Decoder with MediaPipe.
I create 3 categories of body pose with a example. You can add new body pose category for yourself by pose_dataset_create.py.
pose_dataset_create.py - Check the csv dataset is exist or not and with it to add new pose category.
model_train.py - Use pose dataset to train classifier model and save model weights.
model_test.py - Input an image to test the model predict of pose category.
demo_classify_pose.py - Display pose classify and save result with a input video.
Conda virtual env
conda create --name [env_name] python=3.7
conda activate [env_name]
pip install -r requirements.txt
demo_classify_pose.py
python demo_classify_pose.py
If want to demo camera just modify "video_path" set to 0 (or your cameras position number).
And use display_classify_pose() to show result.