一种基于移动机器人视觉的实用数字识别方法  

A mobile robot vision based numeral recognition approach

作　　者：胡春华[1] 马旭东[1] 戴先中[1] 

机构地区：[1]东南大学自动控制系,南京210096

出　　处：《制造业自动化》2006年第5期39-43,共5页Manufacturing Automation

基　　金：国家863计划资助项目(2004AA420110)

摘　　要：移动机器人自动读取数字,并且正确快速识别数字的研究在智能机器人领域具有重要意义。本文以移动机器人为背景,提出一种实时简单的数字识别方法。文中提出边缘图能量百分率法初步定位数字位置。为改善实时识别的准确度,提出改进的中值滤波算法,对采集的数字图片进行去噪处理,利用最大类间方差法分离出目标与背景,采用一种快速、易实现、且对结构损失较少的细化算法,提取数字结构特征,提出分块部分面积法提取特征向量,采用改进的BP网络来设计分类器。试验结果证明这种识别方法不仅快速有效而且稳定。The automatically numerals reading and real-time recognizing of mobile robot are of great significances in mobile robot applications. In this paper, a real-time and simple numeral recognition approach is proposed for application of mobile robot. An energy percentage of edge images algorithm is put forward to localize the numerals. For improving the correct recognition rate, a modified median filter algorithm is employed to remove the image noise, maximum variance between classes algorithm is introduced to segment the object from the background, and a fast, effective and no structures lost algorithm is applied to extract the skeletons of numerals, and then the patch areas as the pattern features vectors are achieved. A modified BP network classifier is designed for accurately numerals recognizing. The experiment results validate that the method is effective, fast and robust.

关 键 词：移动机器人 视觉 数字识别 实时识别 

分 类 号：TP24[自动化与计算机技术—检测技术与自动化装置]