一种基于LBP与HOG联合特征的车标识别方法  

A Vehicle Logo Recognition Method Based on LBP and HOG Features

作　　者：张素雯 张永辉[1] 杨洁[2] 李松斌[2] 

机构地区：[1]海南大学信息科学技术学院,海口570228 [2]中国科学院声学研究所南海研究站,海口570105

出　　处：《网络新媒体技术》2016年第6期49-55,共7页Network New Media Technology

基　　金：国家自然科学基金(61303249);海南省应用技术开发项目(ZDXM2015103)

摘　　要：在基于视频的现代交通监控中,车牌作为车辆的重要信息仍存在识别不准的情况,车标作为车辆的另一个清晰且重要的标志,能够很好的辅助车辆检测。针对多数车标识别方法中难以精准定位、对恶劣环境适应性不强的问题,本文提出一种基于SVM二分类的车标识别方法,选取方向梯度直方图(HOG)和改进的局部二值化特征(LBP)作为特征训练标准支持向量机(SVM)分类器,以一对多的方式实现多分类,将非车标样本也加入到训练负样本中,即使定位稍有偏差也能够正确做出分类,且部署简单,在光照变化和噪声污染的情况下仍能保持较高的准确率,能适应恶劣环境,分类准确率高达99.41%,具有较强的泛化能力。In modern existing intelligent traffic system based on the video, all of the information of the license plate can he captured very accurately. the license plate is of great importance to vehicles, but not The vehicle logo, another clear and important symbol of the vehicle, can be the important auxiliary of vehicle detection. For most of the vehicle logo recognition method, the logo is difficult to be pinpointed, and the recognition is roughly to be done in bad environment. This paper proposes a vehicle logo recognition method based on SVM （support vector machine） classification, firstly, Joint gradient direction histogram （HOG） and the improved local binary pattern （LBP） as the final feature to train SVM classifier, and then carry out the multi -category classification in the form of one to many, which is sample to achieve. The Negative samples which are not the car logo are also added to the training samples to relieve the error caused by slight deviation of the positioning. Experimental results show that the classification accuracy of this method is as high as 99.41% on the original dataset. The proposed method can adapt to harsh environment and has strong generalization ability.

关 键 词：车标识别 HOG特征 LBP特征 SVM分类器 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]