It's badly worded in the article but I think it means person A is entering his pin. Persons B, C, D, etc. take selfies in the vicinity of person A. By comparing multiple selfies from sightly different times, you can determine person A's pin.

I assume the intent is to capture it while surveilling someone while they enter their PIN, not necessarily from images harvested from social media or anything. Since many PIN entries show at most the most recently entered number, you'd need multiple images to capture the PIN. But if you can capture it from reflections in their eyes, then you can surveil them from a greater distance and more stealthily if you've got a good camera.

Reflections of the fingerprint marks on the screen that are usually present for those who don't regularly wipe their screen off?

Take a selfie with someone unlocking their phone positioned over your shoulder. Seems very practical and surreptitious.