田径运动员号码牌图像的号码识别  

Research on the number recognition based on athlete number plate image

作　　者：赵丽科 郑顺义[1,2] 马浩[1] 王晓南[1] 魏海涛[1] 

机构地区：[1]武汉大学遥感信息工程学院,武汉430079 [2]地球空间信息技术协同创新中心,武汉430079

出　　处：《华东师范大学学报（自然科学版）》2017年第3期64-77,86,共15页Journal of East China Normal University(Natural Science)

基　　金：国家863计划项目(2013AA0630905);中央高校基本科研业务费专项资金(2042016kf0012);湖北省科技支撑计划项目(2015BCE080)

摘　　要：田径运动项目中通常拍摄得到大量的图像,如何快速获取特定运动员的图像成为普遍关注的问题.为了快速检索包含特定运动员的图像,本文提出了识别图像中运动员编号的方法,依据运动员编号的识别达到快速检索的目的.首先,采用DPM(Deformable Part Model)(可形变部件模型)进行人体检测,缩小搜索范围,接着按照运动员号码牌的先验知识,采用两种方式进行运动员号码牌定位,保障定位的可靠性;然后对定位出的号码牌进行字符分割;最后采用基于特征的BP(Back Propagation)神经网络的方法进行号码牌识别.实验结果表明,在运动员号码牌几乎无遮挡的情况下,使用本文提出的方法能有效地识别出完整号码牌;在运动员号码牌存在部分遮挡时,可以识别出未被遮挡部分的编号.本文提出的运动员号码牌识别方法为检索特定运动员图像提供了思路,大大减少了普遍采用的人工查找方式的工作量.A lot of images are usually photographed in the sports of track and field, and manual operation is a general method to retrieve image containing certain athletes. In order to quickly retrieve images containing a particular player, a method based on the number to identify a player is proposed. It firstly applied DPM （Deformable Part Model） algorithm to narrow the search scope of number plate. Secondly, according to the prior knowledge, the position of number plate can be located by considering two different ways to ensure the reliability. Thirdly, the characters of number plate are segmented by connected component analysis method. Finally, feature-based BP （Back Propagation） neural network is adopted to recognize the number plate. Experimental results show that the proposed method is efficient to identify the number plate of the players. The number plate recognition method provides a guideline of retrieving a specific player＇s images.

关 键 词：田径运动 字符分割 BP神经网络 字符识别 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]