You're right of course, it can't rely on collimated light like the normal Schlieren system. But the background-oriented Schlieren technique, mentioned elsewhere

You're right of course, it can't rely on collimated light like the normal Schlieren system. But the background-oriented Schlieren technique, mentioned elsewhere in this thread, is more reasonable to look at for inspiration, and it provides pretty much the same end result. I suppose it can be thought of as a differential filter: only minute changes are interpreted to create a representation of the flow of air.