基于先验知识和纹理方向性的车标定位方法  

VEHICLE-LOGO LOCATING METHOD BASED ON PRIOR KNOWLEDGE AND TEXTURE DIRECTIONALITY

作　　者：王建[1] 任明武[1] 

机构地区：[1]南京理工大学计算机科学与技术学院,江苏南京210094

出　　处：《计算机应用与软件》2014年第5期163-167,共5页Computer Applications and Software

摘　　要：基于先验知识,根据车牌与车标的相对位置关系,首先对车标进行粗定位,得到包含车标图案的矩形粗定位区域。为了利用车标背景散热片的纹理方向特征,引入纹理方向测度的概念。根据纹理测度将散热片纹理分为两大类:类水平纹理和垂直纹理。在精定位阶段,对于不同的纹理类型采用不同的定位方法:对于类水平纹理,采用动态模板定位边界,对复杂水平纹理和复杂网状纹理具有良好的定位效果;对于垂直纹理,引入边缘颗粒度的概念,既有效地消除非车标区域的噪声点,又完整地保留车标本身的信息。实验结果表明,该方法鲁棒性好,能够快速、准确地定位车标位置,平均成功定位率为97%,在测试环境下平均定位时间为22ms,定位效率和定位成功率能满足实时系统的需要。Based on the prior knowledge and according to the relative position relationship of vehicle licence plate and vehicle-logo,first we make rough positioning of the vehicle logo and get a rectangular rough location region containing the vehicle-logo image. In order to utilise the texture directional feature of the vehicle-logo background cooling grills,we introduce the concept of texture direction measure. According to the texture measure,the textures of the grille are grouped into two categories： the horizontal-like texture and the vertical texture. In the stage of accurate location,we adopt different location algorithms in light of different texture categories respectively. For horizontal-like texture,we use dynamic templates to locate the boundary,this achieves sound location effects for sophisticated horizontal texture and complicated mesh texture; For vertical texture,we introduce the concept of edge granularity,it effectively removes the noises points outside the vehicle logo region and preserves entire information of the vehicle-logo its own. Experimental results show that the method in this paper has good robustness,it can fast and accurately locate the position of vehicle-logo,the average successful locating rate reaches 97%,and the average locating time is 22 ms in testing environment. This location efficiency and accuracy can meet the demand of real-time system.

关 键 词：车标定位 纹理方向特征 边缘密度 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]