对弈机器人的视觉图像处理和识别  

IMAGE PROCESSING AND RECOGNITION IN THE VISION OF CHINESE CHESS PLAYING ROBOT

作　　者：张志伟[1] 孔凡让[1] 赵吉文[1] 何清波[1] 吴增荣[1] 

机构地区：[1]中国科学技术大学精密机械与精密仪器系,安徽合肥230027

出　　处：《计算机应用与软件》2008年第2期215-217,共3页Computer Applications and Software

摘　　要：基于视觉的对弈机器人在国内尚无先例,其视觉系统的实现是此研究的关键。提出了彩色空间变换,阈值分割、形态学骨架化及霍夫变换等图像处理技术对棋局中的棋子进行检测、定位和分割的方法。在棋子识别过程中,为克服棋子摆放方向的随意性,提取棋子旋转不变的径向像素点数特征;用BP神经网络进行识别,并采用贝叶斯正则化方法提高了网络的推广性。实验表明,该方法定位、分割棋子准确无误,识别率较高。Vision-based Chinese chess playing robot is presented, and the realization of its vision system is the key point for research. Firstly, technologies of color spacial transformation, threshold segmentation, morphological skeleton and Hough transform are applied to detect, locate and segment the chesses in the image. Then, in the process of recognition, the rotational invariant descriptions are extracted, and BP neural networks are used to recognize the characteristics of the chesses. Finally, Bayesian regularization is applied to improve generalization. It is proved by experiments that the method is precise, accurate, and has high recognition rate.

关 键 词：对弈机器人 彩色图像处理 BP神经网络 贝叶斯正则化 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TP242.62[自动化与计算机技术—计算机科学与技术]