The application aims at giving alert on any accidental falls of people living by themselves using human pose estimation. The application uses algorithm from paper Realtime Multi-Person 2D Pose Estimation using Part Affifinity Fields to generate the human pose feature points. And calculate the angle between human body and floor to see whether one person is on the ground for more than five seconds.
Python 3.7
PyTorch==1.7.1
Windows10 20H2
Run pip install -r requirements.txt
Run convert_annotation.py first to generate a new compiled json file.
Put the WFLW dataset folder in the parent folder. Run training.py which will generate a folder models which saves trained model in ckpt.pth.tar.