All you need is to obtain each frame from the video using libraries like OpenCV and parse it into the prediction function.