基于目标轮廓与骨架特征的棋子识别算法

Chess piece recognition algorithm based on target contour and skeleton features

针对中国象棋机器人系统中棋子识别问题,提出了一种基于目标轮廓与骨架特征的棋子识别算法。首先,采用Hough圆检测进行棋子粗定位及预处理。随后,对单幅棋子图像进行形态学处理,提取最大面积轮廓,并利用其最小外接圆进行定位修正。最后,对定位修正后的棋子图像提取其外轮廓与内骨架,计算其Hu矩作为特征向量,并利用支持向量机(SVM)进行识别。以直径为25 mm的棋子为测试对象,利用象棋机器人采集图像进行测试,结果表明,棋子平均识别率在99%以上,平均识别时间为20 ms,完全满足现有象棋机器人需求。

In view of the problem of chess pieces recognition in Chinese chess robot system,a chess pieces recognition algorithm based on the feature of the target contour and the skeleton is proposed. Firstly,the Hough circle detection is used for the rough location and preprocessing of the chessmen. Then,the single piece image morphological processing,extract the maximum area contour,and position correction using the minimum circumscribed circle. Finally,we extract the contour and skeleton of the corrected chessman image,calculate its Hu moments as the eigenvectors,and send them into the trained support vector machine( SVM) multiple classifiers. Using the chess robot developed by us,taking the diameter of 25 mm pieces as the test object,the images were collected and tested. The result shows that the average recognition rate of the chess pieces is above 99%,and the average recognition time is 20 ms,which fully satisfies the needs of the existing chess robots.