This project lets you detect the type and intensity of an action performed by a person in front of a webcam. The model uses PoseNet to prediction the key points of a human in every frame of a video. The values for these key points are then feed into a convolution neural network and a prediction of the action being performed is made.
Once the model is trained it can run live via a webcam.
The datasets for video-action pairs can be collected using the project Action-Recording.
tensorflow 2.3 numpy
wraps a run loop which collects webcam frames and stores the extracted pose information in a person object.
Holds all the information for a person which is extracted from a webcam feed.
Package containing:
- an Predictor ABC
- posepredictor module
- freqpredictor module
- actionpredictor module