With some of my latest experimentation I could 'cheat':

1) Using a FBO, set output to texture A and draw the first object
2) Again with a FBO, set the output to texture B and draw the second object (your poly)
3) Blend the two with difference blending - given the right colours, you get a pattern of where they overlap if they overlap at all.
4) Scan for the pattern and voila - pixel perfect collision detection with openGL Hardware acceleration.

Its 'cheating' in this forum but in my mind should work - my main problem is getting the FBOs to work correctly with blending (almost done 100% accurately) and finding a fast way to scan for said overlap colour (which is why I am reading through a 600 ofdd page book on OpenGL) Just my 2 pence on this problem...