ChessVision: Chess Board and Piece Recognition

This paper details a method to take an image of a chess board and output a reconstructed computer representation of the board through board and piece recognition. Though techniques for board recognition have been thoroughly explored in the past, especially in relation to chessboard calibration, previous works on piece recognition often focus on non-robust segmentation procedures that rely heavily on the exact color of customized chess boards and pieces. The method presented in this paper improves upon the segmentation-based approaches of previous work in piece recognition by introducing a novel approach using classifiers trained on feature descriptors, which is more robust to the similarities in color seen in real-life chess boards. This work is important for both automating the recording of moves in human chess games and improving the ability of chess-playing AI that depend on vision.