This algorithim has three main passes.

 1) 8 by 8 blocks have their average RGB value converted to hue and saturation.
 2) The hue is compared to the tracking colour and modulated by the saturation (saturation = more intense colour)
 3) A cellular automaton is run over the data several times to establish a king of the hill

The 'King of the Hill' pass essentially rewards cells which have neighbours with the target colour. After a few passes the centre of mass of the coloured object will have the highest value. To ensure the tracking is more accurate than 8 pixel cells we then use the weighted position of the 'king' and it's 4 neighbours. The absolute value of the king also gives us an idea of the diameter of the object in question. Finally to provide the benefits of temporal tracking we leave reduced king of the hill values in the grid to lightly weight the next pass.