If you're testing said buttons with a custom machine, don't you already know where the buttons are beforehand? Why would one need to analyze single frames and look for buttons using machine vision?