Automatic Human Action Recognition in a Scene from Visual Inputs