高铁隧道接触网号牌定位与识别  

Location and recognition of catenary number plate in high-speed railway tunnel

作　　者：王凡[1] 张建州[1] 边昂[1] WANG Fan;ZHANG Jian-zhou;BIAN Ang(College of Computer Science,Sichuan University,Chengdu 610065,China)

机构地区：[1]四川大学计算机学院,四川成都610065

出　　处：《计算机工程与设计》2023年第9期2829-2836,共8页Computer Engineering and Design

基　　金：中国博士后科学基金项目(2022M712235);四川大学专职博士后研发基金项目(2022SCU12074)。

摘　　要：为解决高铁隧道接触网图像质量差导致号牌定位困难的问题,提出一种“粗定位-精提取”定位技术。通过二值化标记和特征筛选完成号牌粗定位;对粗定位结果进行边缘补偿,通过二次定位和分类处理实现号牌精确提取,定位准确率达98.4%。为解决低光号牌字符与背景分离难的问题,提出基于多尺度高斯拉普拉斯算子的零交叉二值化方法。在此基础上通过对现有字符分割算法改进,实现字符的正确分割。使用支持向量机和自建数据集进行模型训练和编号识别,识别准确率达95.4%。To solve the problem of the poor image quality of number plate location in high-speed railway tunnel catenary,a rough location-fine extraction location technology was proposed.The rough area of the plate was achieved through binary marking and feature screening.Edge compensation was performed on the rough results,and the plate was accurately extracted through secondary positioning and classification processing,with a positioning accuracy of 98.4%.To solve the problem of separating low-light license plate characters from the background,a zero-crossing binarization method based on multi-scale Gaussian Laplace operator was proposed.On this basis,the correct character segmentation was achieved by improving the existing character segmentation algorithm.Support vector machine and self-built data sets were used for model training and number recognition,and the recognition accuracy is 95.4%.

关 键 词：隧道低光图像 接触网 号牌定位 字符分割 字符识别 二值化 支持向量机 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]