Computer vision system for the chess game reconstruction

In this paper we describe the system for real-time machine vision recognition of chess table and figures. Input data are two synchronized video sequences from a top-view and side-view camera showing the game of chess between two players. The top-view is used mainly for determining the positions of the figures on the table while side-view enables correct recognition of figures in cases where they are not occluded by another figure in front of them. Even in cases where that happens the top-view helps to track the occluded figures along the table. No prior knowledge of the chess rules is used, so there are no assumptions on how and where particular figures are allowed to move. OpenCV algorithm implementations are used for most of the preprocessing and analysis, including the identification of the chess table and classification of figures with SVM classifier.