The title of my project

Detection of facial emotions based on actions

Please watch the below video for a short demo

Procedure : I used the media-pipe holistic model from google, which detects the key points for all the parts of the human body, and then with the help of the utils method, all those key points are drawn, shown using an open cv on the screen. Then we set up folders with names to which gestures we want our model to be trained (In my case, they are "Hello," "Thanks," "I love you"). The key points are collected concerning each gesture and converted into NumPy format, which further is trained using the Long-short-term-memory(LSTM). Finally trained model is used to detect the gestures in real-time.

Dataset : Around 25 - 30 videos are collected for each gesture in 5 sec time frame.

Result : Above machine learning model predicts the facial emotion based on the action performed.