基于非限制场景ALPR系统的车标定位  

Vehicle Logo Location Based on Unconstrained Scenarios ALPR System

作　　者：焦志全 Jiao Zhiquan(School of Mechanical and Automotive Engineering,Shanghai University of Engineering Science,Shanghai 201600,China)

机构地区：[1]上海工程技术大学机械与汽车工程学院,上海市201600

出　　处：《农业装备与车辆工程》2020年第10期72-75,共4页Agricultural Equipment & Vehicle Engineering

摘　　要：现有车标定位技术大多研究正向、具有对称性特点的车辆图像,难以有效解决光线干扰、车辆图像扭曲等复杂应用场景下的车标定位问题。针对该问题,基于非限制场景ALPR系统并结合迁移学习的技术,研究车标定位问题:首先基于ALPR系统精确定位车牌的位置,其次根据车牌和车标的相对位置关系,对车牌位置的4个顶点进行线性变换,最后得到包含车标区域的4个顶点坐标,并依据坐标进行相应透视变换,提取包含车标的区域。在测试集上的测试结果表明:输入正向视角的车辆图像时,定位精度达到99.28%,输入倾斜视角的车辆图像时,定位包含车标区域的准确率为96.56%。Most of the existing vehicle positioning technologies focus on frontal view and symmetrical vehicle images,and it is difficult to effectively solve the problem of vehicle positioning in complex application scenarios such as light interference and vehicle image distortion.In response to this problem,the article uses the unconstrained scenarios ALPR system combined with the technology of migration learning to study the problem of vehicle positioning:firstly,locate the license precisely based on ALPR system;secondly,according to the relative positional relationship between the license plate and the vehicle logo,the four vertices of the license plate position are linearly transformed.Finally,the coordinates of the four vertices including the vehicle logo area are obtained,and corresponding perspective transformation is performed according to the coordinates to extract the region including the vehicle logo.The test results on the test set show that the positioning accuracy is 99.28%when inputting the vehicle image of the frontal view of angle.When the vehicle image with oblique viewing angle is input,the accuracy of positioning the vehicle-containing area is 96.56%.

关 键 词：车标定位 ALPR 复杂场景 相对位置 透视变换 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]