I love this...I play around with OpenCV as much as a Rubyist can (with our kind-of-maintained wrapper) and would learn Python just to make better use of OpenCV...I don't know if what OP is doing is the most efficient way computationally, but it sure is clever, at least for me :)

For instance, recognizing the cards in an image...my first approach would've been to try to identify the numbers/letters of the cards...which would've been just stupid...but the OP's strategy of using findContours and simply choosing the four largest rectangles is elegant...in the way that OpenCV is so powerful you can problem-solve at a fairly high level.