交通标识牌自动识别系统的关键技术的研究

【作者】 于利云；

【导师】 杨忠根；

【作者基本信息】 上海海运学院 ， 通信与信息系统， 2003， 硕士

【摘要】 在全球大力发展智能交通系统的背景下，交通标识牌自动识别系统作为其一个组成部分已经越来越引起学术界和交通部门的注意，在国外其已经成为一个重点研究课题。而在国内由于起步较晚，目前还没有现成的可应用的系统。结合如此现状，适逢摩托罗拉公司又提出了这个课题，因此我们大胆接受挑战，进行交通标识牌自动识别系统的关键技术的研究。 交通标识牌自动识别系统作为一个视觉识别系统，其输入数据是由摄像机所拍的图像，通过识别得到交通标识牌的内容。在本文中，因课题定位输入的图像为静态图像，所以所处理的图像都是静态图像。本文设计交通标识牌自动识别系统由以下几个模块组成：图像分割、摄像机标定、特征曲线的提取、三维复原和图像匹配。因此本文也主要研究这几项技术。 图像分割是任何一个视觉系统的关键技术，是对图象进行处理的基础。在本文中，重点研究了各种阈值分割技术和Canny算子分割方法。因为Canny算子能够提供梯度方向角，而且其分割效果也较为理想，本文采用了Canny算子进行图象分割。摄像机标定是视觉系统不可缺少的一个组成部分，在本文中我们研究了以针孔模型为模型的线性摄像机标定技术，并根据经验确定了简单且可行的摄像机标定技术。特征曲线的提取技术也是一个重点研究的方向，根据交通标识牌多为圆形和多边形的特点，我们研究出了椭圆和直线的提取技术，并研究了优势点提取技术。三维复原是本文中最重要的一个研究点，也是本文中最具创新性的所在。通过研究我们提出了平板圆复原和三角板复原技术，并将它们模块化，使之具有较强的可移植性。图象匹配是物体复原后进行具体识别的一个过程。本文中，我们详细介绍了图像匹配的原理和各种匹配技术。根据交通标识牌的实际情况和本课题的要求，我们按照光学字符识别的原理，通过分类匹配，对图象进行了匹配。 应该说本文的研究立足点在于对静态交通标识牌的识别系统的关键技术的研究上，并没有考虑到系统的实时性。但考虑到我国目前的研究现状，本文的理论研究对于交通标识牌的识别还是有一定的作用的。因此本文的研究是有一定的现实意义的。更多还原

【Abstract】 With the worldwide development of the intelligent traffic system, traffic sign automatic recognition, as a part of the development, is now receiving more and more attention from academe and traffic department. It is an important study project abroad, in China, however, there is no such a traffic sign automatic recognition system available now due to the late start of its development. Considering the actuality and the situation that Motorola has brought forward the plan of recognizing the traffic sign, we meet the challenge and begin to study traffic sign automatic recognition system.As a vision recognition system, the input data of the traffic sign automatic recognition system are images from the camera. By recognizing these images, the system can recognize the content of the traffic sign. In the thesis, images processed are all static because the input images in the project are static. The traffic sign automatic recognition system is composed of the following modules: image segment, camera calibration, characteristic curve extraction, three-dimension recovery and image matching. In the thesis, we mainly study these techniques.Image segment is the key technique for each vision system, and the base of image processing. In the thesis, emphases are put on the study of some threshold-based segment techniques and Canny operator segment method. The thesis selects Canny operator to segment the image because Canny operator can provide the gradient orientation angle and it can segment the image efficiently. Camera calibration technique is an indispensable part of the vision system. In the thesis, many efforts are made on the study of the linear camera calibration technique, which is based on pin-hole imaging model and at last a simple and feasible camera calibration technique is determined according to the experience. Characteristic curve extraction technique is also an important technique. The techniques of ellipse and line extraction, including the dominant point detection are studied in the thesis on the basis of thecharacteristics of the traffic sign. Three-dimension recovery is the most important technique expounded and also serves as the most innovational point in the thesis. Through research, we put forward the techniques of the circle and triangle object recovery and make them as the modules so that they can be easily transplanted. Image matching is a recognition process after the object is three-dimension recovered. In the thesis, we spare no effort to introduce the image matching theory and some matching methods. In light of the optic character recognition principle and after the dominant feature matching, the images are matched to the module images.The image being static, the technique of the system isn’t real-time. But considering the present study situation in china, the theory study in the thesis makes for the advance of the traffic sign recognition. Therefore the thesis has its practical significance.更多还原