基于高斯颜色模型和SVM的交通标志检测  

Traffic sign detection based on Gaussian color model and SVM

作　　者：常发亮[1] 黄翠[1] 刘成云[1] 赵永国[2] 马传峰[1] 

机构地区：[1]山东大学控制科学与工程学院,济南250061 [2]山东省机器人与制造自动化技术重点实验室山东省科学院自动化研究所,济南250061

出　　处：《仪器仪表学报》2014年第1期43-49,共7页Chinese Journal of Scientific Instrument

基　　金：国家自然科学基金(60975025,61273277);山东省自然科学基金(ZR2011FM032,2007ZRA01001);教育部留学回国人员科研启动基金(20101174);山东省机器人与制造自动化技术重点实验室开放课题基金(2011-001);高等学校博士学科点专项科研基金(20130131110038)资助项目

摘　　要：针对我国交通标志的特点,提出一种基于高斯颜色模型和机器学习的快速交通标志检测算法,解决了单纯采用颜色模型或单纯采用分类器检测误差较大的问题。先对直方图修正后的标志图像使用高斯颜色模型分割,并对分割后的图像进行形态学处理,初步提取出候选交通标志,最后将标志的HOG描述子和SVM结合训练出分类器,使用该分类器进行标志的精确检测。实验结果表明,该算法能有效地提高检测精度,降低误检率,对光照、旋转、部分遮挡等不良条件下的交通标志检测具有较优的稳定性和准确性,且满足实时性要求。Aiming at the characteristics of Chinese traffic signs,in this paper, a fast traffic sign detection algorithm based on Gaussian color model and machine learning is proposed. This algorithm solves the problem of low detection rate when only color model or only classifier method is used in traffic sign detection process. Firstly, after the histogram correction, the Gaussian color model method is used to segment the image. Secondly, the morphology processing is uti- lized on the segmented image to extract the candidate traffic signs. Finally, the HOG descriptor and SVM are combined to train the classifier,which is used to detect the traffic sign accurately. Experiment results show that this algorithm can effectively improve the traffic sign detection accuracy, decrease the error detection rate, has strong stability and accuracy for the traffic sign detection in lighting ,rotating and partially occlusion, and also meets the real time requirement.

关 键 词：交通标志检测 高斯颜色模型 HOG描述子 支持向量机 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]