Action Recognition

• Developed a Starting app that can detect only 3 action.
• The App is based on Pose Landmarks Extraction.
• Created a Dataset of 3 action i.e. “Hello”, “Thanks”, “I Love you”.
• Each action contains the 30 video stored as NumPy arrays
• The array contain the Coordinates of the all the Landmarks, especially Hands Landmarks
• I trained a Sequential LSTM model

Dataset

Action : Hello

Hello.mp4

Action : Thank you

Thank.you.mp4

Action : I Love YOU

I.Love.you.mp4

Final App Inference

Action.Recognition.app.mp4

Model Performance (Confusion Matrixs)

Action : Hello	Action : Thanks	Action : I love You