车牌精确定位算法探究

发布时间：2018-05-31 22:47

本文选题：车牌定位 + 形状回归　；参考：《浙江大学》2017年硕士论文

【摘要】：车牌识别系统是现代智能交通系统的重要组成部分之一,被广泛应用于出入控制、车流监控、电子收费等多个场合,提高了交通管理自动化程度。车辆识别系统通过分析和处理复杂背景下的车辆图像,检测、定位车牌,识别汽车牌照的字符,从而快速识别车辆身份。其中车牌精确定位步骤是车牌角度矫正和准确字符分割的重要基础。传统的车牌定位方法利用车牌几何、颜色和纹理特征,处理步骤复杂,适用场景有限,在低照度、透视变换、低质模糊等场景下的准确率有待提高。为了构建更加通用准确的车牌精确定位算法,本文受人脸关键点检测方法的启发,采用形状回归方法,将车牌的精确定位转化为求取车牌四角坐标。借助大量标注后的车牌数据,学习车牌的角点特征,建立多个阶段的回归方程,通过每一阶段的反馈调整,逐渐接近真实的车牌位置。通过在真实复杂的车牌数据集里进行实验,我们证明基于形状回归的车牌定位算法有着更快的定位速度,和更低的位置偏移,对车牌的拍摄环境、拍摄角度和距离依赖较小,具备更高的通用性。同时,基于形状回归的车牌定位算法能够提高车牌识别的准确率。
[Abstract]:License plate recognition system is one of the important parts of modern intelligent transportation system. It is widely used in many occasions such as access control, vehicle flow monitoring, electronic charge and so on, which improves the degree of automation of traffic management. The vehicle recognition system can quickly identify the vehicle identity by analyzing and processing the vehicle image under the complex background, detecting, locating the license plate and recognizing the characters of the vehicle license plate. The accurate location of license plate is an important basis for the correction of license plate angle and accurate character segmentation. The traditional license plate location method uses the license plate geometry, color and texture features, the processing steps are complex, the applicable scene is limited, and the accuracy of low illumination, perspective transformation and low quality fuzzy scene needs to be improved. In order to construct a more general and accurate license plate accurate location algorithm, this paper, inspired by the face key point detection method, uses the shape regression method to transform the accurate license plate location into obtaining the license plate quadrangle coordinates. With the help of a large number of tagged license plate data, learning the corner feature of license plate, the regression equation of multiple stages is established, and through the feedback adjustment of each stage, the real license plate position is gradually approached. Through experiments in the real and complex license plate data set, we prove that the shape regression based license plate localization algorithm has faster localization speed and lower position offset, and has less dependence on the shooting environment, shooting angle and distance of the license plate. It has higher generality. At the same time, the algorithm based on shape regression can improve the accuracy of license plate recognition.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：U495;TP391.41

【相似文献】