We changed our existing design from a centralized dealing mechanism to users inputting the tiles they are dealt. This way, we only need one camera for board observation. Also, we then mitigate the risk of incorrect letter recognition because we can check that new tiles recognized on the board were indeed in the player’s rack. Before, we were relying on two instances of letter recognition by our computer vision. Now, we have a method for increasing our confidence that the letter recognition was correct.