Methods for location and recognition of chess pieces based on machine vision

Abstract

The positioning and recognition of chess pieces is the key to the chess robot's recognition of games. Aiming at the positioning and identification of chess pieces a method of centroid positioning of the maximum connected components and a method of convolutional neural network identification were proposed. First, the pieces were pre-segmented based on the HSV color space, and then the segmentation results were eroded and dilated to obtain the maximum connected components of the pieces, and the centroid coordinates of the connected components were calculated as the positioning results of the pieces. Finally the chess pieces were identified using the trained convolutional neural network. The network used a variety of techniques to reduce the number of parameters under the premise of ensuring a high recognition rate so that it could be deployed on resource-constrained devices. The results show that the average positioning error of the method on a chessboard with a size of 28cm and 28cm is 0.48mm, the average positioning time is 11ms and the recognition accuracy of chess pieces is 98.7%•