Edited post with some example pics. I've been leaning toward #3 if I can't find a better solution, but can you provide more info about #2? My labels are the (x,y) coordinates of each corner of the cards.

Thanks for the suggestions. I edited my post to give some examples of the detection that needs to be performed... notice how sometimes corners can be obscured, and the background can make "simple" rectangle detection a poor fit. I will check out ORB though.

For this particular problem, the image could contain many corners - and even other full rectangles. The goal is to detect the specific type of paper card I'm interested in - easily identifiable based on color/pattern - but not easily extracted from a Sobel filter.