One of the problems with traditional pattern tracking systems is they are configured to track a single pattern in a 2d scene. While these individual patterns can be combined into groups to form more complex shapes the patterns are still discreet. This requires encoding their position within the larger pattern in a higher level structure. They are also sensitive to partial occlusion because they are designed to have minimum reducnancy to give best recognition from the generally low quality, low resolution web cameras.

The new fudicial system I am working on provides continuous patterns which also encode a sequential counter in x and y axis. This pattern is designed to be read even when large chunks are occluded. Finally the 3D scene matching will take advantage of the higher level primitives such as cylinders, strips and grids to provide more accurate tracking.