车牌检测级联分类器快速训练算法  

Improved license plate detection method based on AdaBoost algorithm

作　　者：方义秋[1] 卢道兵[1] 葛君伟[1] 

机构地区：[1]重庆邮电大学GIS研究所,重庆400065

出　　处：《重庆邮电大学学报（自然科学版）》2010年第1期103-107,共5页Journal of Chongqing University of Posts and Telecommunications(Natural Science Edition)

基　　金：重庆市教委资助项目(KJ090519)

摘　　要：针对传统AdaBoost算法的不足,分析了训练过程中出现过训练及分类器退化的问题,并提出了解决这一问题的有效新方法。新方法主要对样本及时更新和样本权重的更新规则进行了调整。使用该方法训练级联车牌检测器,实验结果表明,新方法较好地解决了传统AdaBoost算法中所出现的过训练及退化问题,在保证检测率的同时降低了误检率,并且训练时间缩短了50%左右。Focusing on the disadvantages of classical AdaBoost algorithm, this paper analyzes the issues of excessive training and overfitting for classifiers and proposes a new method to avoid these problems. The new method is to update the training samples in time and regulate the updated rules of sample weights. As a result, using the method to train a cascade license plate, the experimental results show that the new method does not lead to the issues of excessive training and overfitting like classical AdaBoost often does, and moreover, it will reduce false alarm rate with a high detection rate and the training time is shortened to 50%.

关 键 词：ADABOOST算法 样本更新 权重调整 车牌检测 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]