【作者】 李秋洁； 茅耀斌； 王执铨； 项文波；

【Author】 LI Qiu-jie MAO Yao-bin WANG Zhi-quan XIANG Wen-bo(School of Automation of Nanjing University of Science & Technology,Nanjing 210094)

【机构】 南京理工大学自动化学院；

【摘要】 车牌识别是智能交通、安防监控领域的热点问题之一,车牌检测是车牌识别系统的第一步,由于受到光照条件、拍摄视角、背景噪声、运动模糊等因素的影响,复杂场景下的车牌检测成为一个难题。Adaboost 算法作为一种有效的学习方法已成功运用到人脸检测中,本文设计了一种基于 CS Adaboost 的车牌检测算法,主要贡献在两个方面:1.改进现有的用于 Adaboost 训练的车牌特征提取方法;2.提出一种基于 CS Adaboost 的代价因子引入方式,加快了训练过程的收敛速度,提高了分类器的泛化能力。最终得到的车牌检测器在测试集上取得了较好的结果。更多还原

【Abstract】 License Plate Recognition(LPR)is a pop technology in intelligent transportation and safety surveillance.As the first step of LPR systems,license plate detection in cluttered background is a difficult task due to many effective factors like variant illumination,different visual angle,background noise,and motion blur.Adaboost algorithm is an effective learning algorithm which has been successfully used in face detection.This paper designs a license plate detection algorithm using Cost-sensitive Adaboost.There are two main contributions:(1)the improvement of feature selection methods for license plate detection;(2)a new method of introducing the cost factor used in Cost-sensitive Adaboost algorithm,which improves convergence and generalization ability.The final license plate detector gets encouraging results in the experiments.更多还原