自然环境下道路交通标志的检测与识别研究

发布时间：2018-05-17 16:50

本文选题：交通标志识别 + 交通标志检测　；参考：《南京理工大学》2014年硕士论文

【摘要】：交通标志识别是智能交通系统(ITS)的重点研究方向之一,该技术可以应用到无人驾驶车辆和驾驶员辅助系统,为自动或半自动驾驶车辆提供有用的道路信息。经过国内外学者几十年的研究,交通标志识别领域的理论和实践体系逐渐形成,并取得了很多突破性的进展。本文主要针对交通标志识别中的检测、特征提取和分类方法进行了研究。在交通标志检测阶段,针对交通标志的颜色和形状特点,本文提出了一种基于颜色分割和局部Hough变换的交通标志检测方法,首先对交通标志图像进行颜色分割,对分割得到的二值图像进行轮廓跟踪提取候选区域,然后依据候选区域的RGB均值对其进行形状预分类,接着运用局部Hough对归一化的候选区域进行形状的检测,最后定位交通标志区域。在进行特征提取时,主要研究了局部核Fisher鉴别分析,分别对基于子模式的核Fisher鉴别分析(Sp-KFDA)口基于模块的核Fisher鉴别分析(MKFDA)进行了分析。结合交通标志信息分布的特点,提出了一种基于自适应加权的模块核Fisher鉴别分析(Aw-MKFDA)进行交通标志识别,通过在K近邻分类器上的比较实验表明,本文提出的Aw-MKFDA方法比Sp-KFDA方法和MKFDA方法具有更高的识别率。通过对分类错误样本的分析,我们发现相似类是导致分类错误的一个重要原因。为了解决由于相似性引起的误分类问题,本文提出了基于相似类划分的两阶段交通标志识别。该方法将交通标志识别过程分为两个阶段：第一阶段用稀疏表示进行相似类的大类识别；第二阶段用稀疏表示进行相似类里的具体类别识别。在稀疏方法用于交通标志识别的过程中,本文提出采用局部字典代替常用的全局字典,解决了交通标志大样本引起的字典过大问题。实验结果表明,本文提出的基于相似类划分的两阶段交通标志识别方法能够有效的提高交通标志的识别率。最后,本文采用无人驾驶平台实时采集的交通标志场景图像进行了综合实验,对本文提出的检测和识别方法进行了验证,并将稀疏表示和局部KFDA的识别方法进行组合,提出了一种基于投票的组合方法。实验结果表明,本文提出的方法获得了比较理想的结果,并且具有一定的稳定性。
[Abstract]:Traffic sign recognition is one of the key research directions of Intelligent Transportation system (ITS). This technology can be applied to driverless vehicles and driver-assisted systems and provide useful road information for autonomous or semi-autonomous vehicles. After decades of research by domestic and foreign scholars, the theory and practice system of traffic sign recognition has gradually formed, and has made a lot of breakthrough progress. In this paper, the detection, feature extraction and classification of traffic sign recognition are studied. In the phase of traffic sign detection, according to the characteristics of color and shape of traffic sign, this paper presents a method of traffic sign detection based on color segmentation and local Hough transform. Firstly, the color segmentation of traffic sign image is carried out. The binary image is extracted by contour tracking, then the candidate regions are pre-classified according to the RGB mean of candidate regions, and then the normalized candidate regions are detected by local Hough. Finally, locate the traffic sign area. In feature extraction, the local kernel Fisher discriminant analysis is mainly studied, and the kernel Fisher discriminant analysis based on the kernel Fisher discriminant analysis (SP-KFDAA) based on sub-pattern is analyzed respectively. Based on the characteristics of traffic sign information distribution, an adaptive weighted modular kernel Fisher discriminant analysis (Aw-M-MKFDA) is proposed for traffic sign recognition. The Aw-MKFDA method proposed in this paper has a higher recognition rate than the Sp-KFDA method and the MKFDA method. Through the analysis of classification error samples, we find that similar classes are an important cause of classification errors. In order to solve the misclassification problem caused by similarity, this paper proposes a two-stage traffic sign recognition based on similarity classification. The method divides the traffic sign recognition process into two stages: in the first stage, the sparse representation is used to identify the large classes of similar classes; in the second stage, the sparse representation is used to identify the specific classes in the similar classes. In the process of sparse method used in traffic sign recognition, a local dictionary is proposed to replace the common global dictionary, which solves the problem of excessive dictionary size caused by large sample of traffic signs. The experimental results show that the proposed two-stage traffic sign recognition method based on similar class partition can effectively improve the recognition rate of traffic signs. Finally, the scene images of traffic signs collected by driverless platform are synthesized, and the detection and recognition methods proposed in this paper are verified, and the sparse representation is combined with the recognition method of local KFDA. A combination method based on voting is proposed. The experimental results show that the proposed method has better results and has certain stability.
【学位授予单位】：南京理工大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP391.41;U495

【参考文献】