基于SURF的多阶段车牌定位算法  

Multi-stage License Plate Location Algorithm Based on SURF

作　　者：孙卓婷 王福龙[1] SUN Zhuo-Ting;WANG Fu-Long(School of Mathematics and Statistics,Guangdong University of Technology,Guangzhou 510630,China)

机构地区：[1]广东工业大学数学与统计学院,广东广州510630

出　　处：《软件导刊》2022年第8期177-182,共6页Software Guide

摘　　要：不同类型车辆的车牌形状、大小及颜色有所不同。为对不同拍摄视角、尺度、背景、光照强度及各种形式遮挡下的各型车牌进行准确检测,提出一种基于SURF算法的多阶段车牌定位模型。该模型考虑车牌丰富的纹理和结构信息,借助SURF特征矩阵的行协方差系数分布定义车牌候选区域的特征,从而得到多个差异明显的分块区域;同时提出一个新的四维特征描述符精准提取车牌候选区域,并基于Hessian矩阵对车牌字符结构特征的度量实现对车牌区域的判别。通过在CCPD数据集上进行测试,发现该模型不需要任何受控条件或环境参数设置,具有定位变形、模糊、污损以及光照变化情况下车牌的能力。The shape,size and color of license plates of different types of vehicles vary. In order to accurately detect license plates under different shooting views,scales,backgrounds,light intensity and various forms of occlusion,a multi-stage license plate localization model based on SURF algorithm is proposed. The model considered the rich texture and structure information of the license plate,using the row covariance coefficient distribution of the SURF feature matrix to define the characteristics of the license plate row candidate area,so as to obtain multiple block areas with obvious differences,and propose a new four-dimensional feature descriptor for precise extraction of the candidate area of the license plate row,finally,the identification of the license plate area is realized based on the Hessian matrix to measure the structural characteristics of the license plate characters. After testing on the CCPD data set,the model does not require any controlled conditions or environmental settings,and has the ability to deal with license plate deformation,blurring,dirtying,and lighting changes.

关 键 词：SURF算法 车牌定位 特征描述子 计算机视觉 智能交通 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]