It seems like you can avoid having to recognize the pieces altogether if you are taking photos of each move during a match, since they have a known starting pos

It seems like you can avoid having to recognize the pieces altogether if you are taking photos of each move during a match, since they have a known starting position. Edit: what I mean is that in this case the problem becomes answering "does this place on the board have a piece on it" not "which piece is on this place on the board." Once you know which places on the board have any pieces whatsoever you can simply infer between moves which pieces moved where.