【作者】 刘红；

【导师】 高向东；

【作者基本信息】 郑州大学 ， 机械电子工程， 2011， 硕士

【摘要】 随着社会进步和经济的发展,我国的公路交通行业得到了持续、快速地发展。高度发达的现代交通为人类的生活带来了便利,但同时交通安全、交通拥挤等问题也变得越来越严重。为了解决这些问题,智能交通系统ITS (Intelligent Traffic System)这一研究领域应运而生。道路交通标志识别系统TSR (Traffic Signs Recognition)作为智能交通系统研究方向,已成为国内外学者研究的热点之一,它通过安装在机动车辆上的摄像机摄取自然场景图像,并将图像送至系统的图像处理模块进行图像理解、交通标志检测与识别,最后将识别结果告知驾驶员,以达到增强道路交通安全、降低交通拥挤的目的。道路交通标志中,警告标志、禁令标志和指示标志是三种最重要、也是最常见的交通标志,它们均具有特定的颜色和形状用以区分其他物体,以达到提醒驾驶员或行人的目的。十几年以来,道路交通标志识别研究有了很好的进展,并取得了一定的研究成果,但因背景复杂性以及光照等各种影响因素的存在,导致了它比非自然场景下的目标识别更具有挑战性,其影响因素主要表现在以下方面：光照条件时常变换且不可控、车辆震动导致摄取的图像模糊、交通标志被损坏、被污染或被遮挡、交通标志颜色褪色、雨雾等恶劣天气的存在,以及投影失真、尺度变换、倾斜、相同颜色背景等。从交通标志的颜色信息和形状特征出发,研究一种交通标志的智能检测算法。该算法主要包括基于HSV(Hue-Saturation-Values)颜色空间的交通标志图像分割和基于颜色与形状的交通标志检测两部分。首先将RGB(Red-Green-Blue)图像转换至受光照影响较小的HSV颜色空间,通过提取不同颜色的阈值范围来定位目标区域；再根据目标区域的几何形状来划分警告、禁令和指示三种不同类别的交通标志,完成交通标志图像的检测。针对自然场景下影响交通标志检测效果的不利因素,研究了基于多尺度Retinex的交通标志图像增强和仿射变换的三角形交通标志校正算法,以及规范化圆形交通标志和矩形交通标志的算法。实验结果表明,该智能检测算法能克服光照、图像模糊、雨雾天气、尺度变换等多种不利因素带来的影响,具有较高的鲁棒性,验证了算法的有效性。支持向量机是一种新的自学习算法,它建立在统计学习理论和结构风险最小原理之上,在小样本的模式识别(分类)中具有很好的优势。交通标志的分类和识别总是在有限个样本中进行的,本文基于支持向量机对已检测出的交通标志图像进行识别研究。挖掘能代表不同交通标志特征的数据：Hu不变矩和Zernike不变矩,并比较C-SVM和v-SVM两种支持向量机分别在Linear核函数、Polynomial核函数、RBF核函数和Sigmoid核函数下的交通标志分类准确率,并在此基础上调整核函数的惩罚因子至最优。通过对不同限速的交通标志图像分类和识别的实验表明,特征数据经归一化、核函数经寻优之后分类和识别均取得了良好的效果。更多还原

【Abstract】 With the development of society and economy, the road transport industry in China has been developed continuously and rapidly. Highly developed modern transportation provides the convenience to human, however, a series of issues, including the traffic safety and traffic congestion, are becoming increasing serious. In order to solve these problems, the Intelligent Traffic System(ITS) Application is come into being, and the road Traffic Sign Recognition system(TSR), as a sub-topic of ITS research field, has become a focus to domestic and foreign scholars. The TSR system can be descript as follows:The natural scene images shot with a camera mounted on a vehicle, are sent to the image processing module for image-understanding, traffic sign detection and recognition, then informs drivers of the final results, as a result, the traffic safety can be enhanced and the traffic jams can be reduced.The warning signs, the prohibition signs, and the directional signs are the most important and common types, there are specific colors and shapes to distinguish them from other objects and remind drivers or pedestrians. More than a decade, the research of road Traffic Sign Recognition has been progressed highly and achieved some results, but the complexity of the background, light, and so on, led to its research more challengeable than the target-recognition in non-natural scenes. The factors are mainly as follows:light-condition and its uncontrollable feature, image-blurring because of vehicle vibration, damaged, contaminated or blocked, faded, weather of rain and fog.etc, projection distortion, scale transformation, titled, same or similar color-background.An intelligent algorithm of detection for the traffic signs based on the color information and the shape features was proposed. It consists of two parts, the first one is the image-segmentation for traffic signs with the HSV(Hue-Saturation-Value) color space, and the other one is the detection for traffic signs using the color information and the shape features. In this algorithm, RGB(Red-Green-Blue) images were converted into HSV color space, then the Region of Interests(ROIs) were located by extracting the thresholds of different types of colors. Based the geometry of ROIs, then, warning signs, prohibition signs and directional signs were divided, in consequence, traffic sign detection was done. In order to overcome several of unfavorable factors which are generally exist in the natural scenes, an image enhancement algorithm based on multi-scale Retinex and two correction approaches for the traffic signs are proposed. The first correction approach is the Affine transformation for the triangular traffic signs, and the second one is the normalization for the circle traffic signs and the rectangular traffic signs. Experiments were conducted for traffic sign detection and the results indicated that this intelligent algorithm could overcome types of unfavorable factors and posses of good robustness, which verified its effectiveness.Support Vector Machine(SVM) is a new self-learning algorithm, which built on the Statistical Leaning Theory and the Structural Risk Minimization principle, it shows good advantages in small-sample-pattern-recognition(classification). Recognition and classification of traffic signs are always conducted in limited samples, and taking this into consideration, it presents an approach for traffic signs recognition based on SVM. Hu-Invariant-Moments and Zernike-Invariant-Moments represent different feature data of different traffic signs, and it is compared that the accuracy of traffic signs classification and recognition with two Support Vector Machines:C-SVM and v-SVM in different types of kernel functions, Linear kernel, Polynomial kernel, RBF kernel and Sigmoid kernel. And for every kernel function, its penalty-factor was adjusted to the optimal. The different Limited-Speed traffic signs experiments were conducted for traffic sign classification and recognition after feature data normalization and kernel function optimization, and the good results achieved.更多还原