您的位置:首页  > 论文页面

基于CTPN的列车车号的定位与识别

发表时间:2019-06-28  浏览量:1525  下载量:215
全部作者: 蔚诗瑶,邹琪,罗常津,曲健聪
作者单位: 北京交通大学计算机与信息技术学院;中国铁道科学研究所电子所
摘 要: 研究提出一种两阶段的车号定位方法,首先基于连接文本建议网络(connectionist text proposal network,CTPN)提取候选车号区域,然后通过非车号抑制方法筛选出唯一的车号区域。分割阶段提出一种动态窗口分割机制。最后融合不同方向、不同视角的分割识别结果。本文在所构建的大规模数据集上进行实验测试分析,验证了所构建的列车车号识别系统在定位、分割、识别以及整体识别各方面均取得鲁棒效果。
关 键 词: 计算机应用;列车车号的定位与识别;连接文本建议网络(CTPN);动态窗口分割;融合
Title: Localization and recognition of train number based on CTPN
Author: YU Shiyao, ZOU Qi, LUO Changjin, QU Jiancong
Organization: School of Computer and Information Technology, Beijing Jiaotong University; Institute of Electronic, China Academy of Railway Sciences
Abstract: A two-stage method for train number location is presented in this paper. Firstly, the train number region candidates are obtained by extracting text regions based on connectionist text proposal network (CTPN). Then, the only train number region is screened out with a non-train number region suppression method. In the segmentation stage, a dynamic window segmentation mechanism is proposed. Finally, the segmentation recognition results of different directions and perspectives are fused to get the final output. In this paper, experimental test and analysis are carried out on the constructed large-scale data set. The experimental results verify that the train number recognition system constructed in this paper achieves robust effects in localization, segmentation and recognition as well as the overall recognition.
Key words: computer applications; localization and recognition of train number; connectionist text proposal network (CTPN); dynamic window segmentation; fusing
发表期数: 2019年6月第3期
引用格式: 蔚诗瑶,邹琪,罗常津,等. 基于CTPN的列车车号的定位与识别[J]. 中国科技论文在线精品论文,2019,12(3):405-413.
 
5 评论数 0
暂无评论
友情链接