So that's how they did it. I thought they had some amazing image and visual recognition. I've been working on a little something similar but these types of problems are hard.