Slap a depth-sensing thing (like Project Tango https://www.google.com/atap/project-tango/) onto the AR stuff, you can probably get a Kinect-like algorithm to sense peoples' limbs, bodies, etc., and come up with a hypothetical-location-in-3d-space. Then the hilarity will start with projecting cartoon body overlays. Successive versions of the app will do better recognition and fitting and come up with more realistic images. Eventually you can have AR "beer goggles" or whatever effect you want.
Also you'll be really sorry when you look at the wrong person.