Got some (not much) time to play around with this again...
Obviously I am not the right person to do that. On the workbench it works fine, but when I get around to implement it, it's problems, problems, problems...
One of them is that lighting is a bit tricky - I have fitted two LEDs to light the area evenly, however, due to restricted space and surrounding (i.e. white glossy paint of the bike frame) it's hard to get good contrast - either two much light is reflected off the paper or it is too dim.
Anyway, I have started a bit too ambitious - I've put the encoding wheel with 24 sections (12 impulses per turn), but there was too much noise to establish the threshold value right. Now I've fitted 8-section wheel - this works well, but then the resolution is the measly 4 impulses per turn... Hall sensors look attractive again, but then I'd need to order the magnets. By the way, anyone has any experience with hall sensors and neodymium magnets? What's the typical distance they detect the field from? I do not want to space them too densely...