Does anyone know how the detection of sudokus on vertical planes can be achieved. Great article esp on the crowdsourcing and machine vision fronts, but the authors explanation of this aspect left a lot to be desired.
The data I had available to mess with was the difference in width of the top of the puzzle and the bottom (with some trig you can determine its angle relative to the camera) and the projection matrix of the camera relative to the scene origin.
It's not perfect but it works better than having nothing at all.