摘要
交通标志包含重要的交通信息,交通标志的自动识别是智能车辆和智能辅助驾驶系统(advanced driver assistance systems,ADAS)的关键技术,快速和准确地识别交通标志对于安全驾驶意义重大。研究基于ORB算法全局特征的交通标志快速识别算法,ORB是一种局部描述算子,通常用于局部特征的匹配,首先将训练的交通标志图像大小归一化,图像的中心点作为ORB的特征点,通过运用归一化的图像大小作为一个ORB图像块,计算了特征点的描述符并且用它作为全局描述符来进行交通标志的识别。识别阶段,在海明距离的基础上利用最近邻域算法识别出交通标志的类型。最近邻算法(K-Nearest Neighbor,KNN)的优点是所有成功识别出来的交通标志可以作为训练资源被训练数据库利用。利用公开的德国交通标志数据集(GTSRB)对所研究的算法进行测试,算法识别准确率达到91%。一个交通标志的特征提取和识别的平均时间小于2ms。实验结果表明,基于ORB算法全局特征的交通标志快速识别算法能精准、高效地识别交通标志,且鲁棒性好。
Traffic signs contain important traffic information.Automatic recognition of traffic signs is a crucial technology for intelligent vehicles and advanced driver assistance systems（ADAS）.Fast and accurate sign recognition is especially important for safe driving.This paper proposes a fast sign recognition algorithm based on holistic features extracted through an Oriented FAST Rotated BRIEF（ORB）method.ORB is a local description operator,which is commonly used for local feature matching.In this paper,an input sign image is first normalized and the central point of the normalized image is set as the location of an ORB feature point.Regarding the normalized image as an ORB patch,the ORB feature point descriptor is computed as sign holistic feature.In the recognition stage,the K-Nearest Neighbor（KNN）method is utilized to recognize sign types based on Hamming distances.One advantage of KNN method is that all the signs recognized successfully can be further utilized as the samples in the training database.This proposed algorithm has been tested with public GTSRB traffic sign database.The recognition rate is 91%,and the average processing time including ORB feature extraction and sign recognition is less than 2ms.Experimental results show that this proposed algorithm is fast,accurate,and robust for traffic sign recognition.
出处
《交通信息与安全》
2016年第1期23-29,共7页
Journal of Transport Information and Safety
基金
国家自然科学基金项目(51208168
51578432)
道路交通安全公安部重点实验室开放基金项目(2015ZDSYSKFKT04)
中央高校基本科研业务费专项资金项目(2014-IV-068)
武汉市青年科技晨光计划项目(2015070404010196)
湖北省自然科学基金项目(2015CFB252)资助
