Computer Engineering and Applications ›› 2023, Vol. 59 ›› Issue (8): 175-181.DOI: 10.3778/j.issn.1002-8331.2112-0246

• Pattern Recognition and Artificial Intelligence • Previous Articles     Next Articles

Distance Posts Detection and Character Sequences Recognition Method in Video Images Acquired from Camera in Moving Vehicle

LIU Xiaoxi, CHENG Jiacheng, CHENG Yongmei, GU Yifan, LEI Xinhua, WANG Bo   

  1. 1.Department of Navigation, AVIC Xi’an Flight Automatic Control Research Institute, Xi’an 710076, China
    2.School of Automation, Northwestern Polytechnical University, Xi’an 710114, China
    3.Shanghai Electromechanical Engineering Research Institute, Shanghai 201109, China
  • Online:2023-04-15 Published:2023-04-15



  1. 1.中国航空工业集团公司西安飞行自动控制研究所 导航部,西安 710076
    2.西北工业大学 自动化学院,西安 710114
    3.上海机电工程研究所,上海 201109

Abstract: The problem of vehicle location in GPS denied environments can be solved with the video images positioning method, which detects the highway distance posts and recognizes characters to obtain their geographic information through information retrieval. In this paper, resulting from few types of distance posts, it puts forward S-YOLOv3(simplified YOLOv3) to calculate more fastly. In order to improve the accuracy of character recognition, it proposes HPCC-YOLOv3 (high precision character classification YOLOv3). S-YOLOv3 and HPCC-YOLOv3 are trained and tested respectively. Characters are clustered according to position in kilometer posts and hectometer posts to realize character recognition. An images acquisition, highway distance posts detection and character recognition system composed of the Daheng mercury camera and a computer is designed. With images acquired from Daheng mercury camera in the experimental car, it gets the results showing that the system performs well, which can effectively improve the speed of highway distance posts detection and the accuracy of character recognition in video images.

摘要: 视频图像定位方法可以解决GPS拒止环境下行车定位问题,该方法对公路路牌检测与字符序列识别,通过地理信息检索,得到路牌所在位置的地理信息。针对路牌检测类别少的问题,对YOLOv3进行轻量化改造,提出了简化的YOLOv3(simplified YOLOv3,S-YOLOv3);为了提高字符分类精度,对YOLOv3进行特征融合策略改进,提出高精度的字符分类YOLOv3(high precision character classification YOLOv3,HPCC-YOLOv3);分别对S-YOLOv3与HPCC-YOLOv3进行训练与测试;按照字符检测结果所处的位置进行字符聚类,实现字符序列识别;设计了由车载大恒水星相机、计算机组成的图像采集、车牌检测与字符识别系统;在复杂环境下进行跑车实验,结果表明了提出的方法能够有效提高视频图像路牌目标检测的速度和字符序列识别的精度。