六子棋自動棋譜產生系統

Abstract

我們提出了一套自動棋譜產生系統，只要在下棋時利用攝影機拍攝棋盤，便可經由程式自動記錄每一步棋的位置及思考時間等資訊，而且只需要一般的攝影機及筆記型電腦即可執行此自動棋譜產生系統。我們先利用色彩資訊找出影像中的棋子區域，再找出影片中的關鍵影格，以及每一手新增之棋子。而在棋子的定位上我們提出了兩種方法，一種是「相對於棋盤之棋子定位」，另一種是「基於棋子相對位置之定位」。其中前者是利用投影轉換矩陣（Homography）把棋子的影像座標轉換至棋盤的座標；而後者則是先建立棋盤上所有棋子的相對關係，再藉此資訊重建棋盤的座標系，如此可以免除偵測棋盤格線所造成的誤差，為一種創新且更具彈性的作法。不管是哪一種定位方式，都可以正確定位每一手棋子的位置，達到即時記錄棋譜的目的，且本系統並不需要限制輸入的影片一定是俯視拍攝的，也針對在比賽過程中棋盤或攝影機可能被移動，而造成分析結果不正確的問題做了完善的處理，更增加了本系統在使用上的方便性。

In this thesis, a system which can automatically extracts the game record, including each stone’s position, color and ply interval, from video of “Connect Six” or “Go” is developed. Firstly, the system will identify all stone regions of each image frame by the color information so that the key frames and incremental stones of each ply can be detected. After identifying stone positions in the image, two approaches can be adopted to find their locations in the real world: (i) find them with respect to board by homographic transformation (as done in some previous works), and (ii) find them by reconstructing a coordinate system of the board based on relative positions among the stones in the image space but without using the board information (as proposed in this thesis). The proposed approach is not only more flexible, i.e., with less constraints imposed on the game video, but also more robust with respect to non-ideal image quality associated with the board image. Experimental results show that both methods can find stone location correctly and generate game record efficiently. Finally, the proposed system is easy to use since the system input is not restricted to a top-view video, and the problem of board movement can also be resolved automaticall