Tensor voting, hough transform and SVM integrated in chess playing robot

Abstract

Robot which replaces and assumes the role of the human is becoming a common and popular problem. Also, game playing robot has challenges to researchers as well as people being interested in this field. In this paper, we want to introduce a new method to detect and recognize chess pieces of Janggi Chess game. Paper is a uniform approach from input image receiving to chess piece recognition. Besides, some new algorithms are used to get the highest performance of system and can apply for real robot system. The first, Tensor Voting is applied to find four corners of chessboard which can extract the full chessboard from background and noise for both simple and complex cases, which other methods are difficult to overcome. Secondly, Circle Hough Transform can detect the chess pieces' size and position correctly regardless of the effects of light, capture angle, the quality of images, etc. Furthermore, the piece recognition step is implemented using SVM (Support Vector Machine), a popular algorithm for classifying with highest performance. The promising results have confirmed the effectiveness of the proposed method.