【作者】 徐永胜；

【导师】 曹洁；

【作者基本信息】 兰州理工大学 ， 控制理论与控制工程， 2007， 硕士

【摘要】 交通信息采集与处理是实现城市交通智能化的关键，是智能交通中各子系统实施的核心基础和重要前提，在交通规划和交通管理控制等方面具有广泛的应用前景。本文利用数字图像处理技术对智能交通中的交通参数检测进行了研究。基于图像识别技术的交通参数检测就是利用数字图像技术，获得被检测车道上车型、车速、车流量、车牌等交通参数，这些车辆信息为智能交通系统的应用提供了有力的保障。交通参数检测涉及面较广，本文仅从车牌识别、车型检测这两个方面对图像信息交通参数检测进行了研究。在车牌识别方面，首先采用边缘检测与行扫描相结合的方法，实现车牌快速准确定位；利用大津法获取最佳阈值实现车牌图像的二值化；同时采用Radon变换、边框去除、形态学处理等算法，实现了车牌精确定位，并从中分割出车牌字符；最后，采用网格法提取字符特征，用BP神经网络实现车牌字符的识别，从而得到车辆的牌照信息。在车型方面，采用自适应背景生成与更新算法，获取道路交通现场的背景，并采用减背景法实现车辆目标的检测；采用二值化、边缘检测、形态学处理来获得车辆外形特征；此外，考虑到受外界环境的影响以及传输干扰而造成的检测精度低，本文采用D—S证据理论来对多源图像数据进行融合处理，从而获得相对准确的车型参数。论文详细论述了本系统的设计思想、特点、各组成部分的构成和功能，完成了车牌识别与车型检测的软件设计，对设计中关键技术进行了详细的介绍。更多还原

【Abstract】 Traffic information acquisition and processing is the key to urban intelligentized transportation, and has wide applications in transportation planning and transportation manage control utilities. This paper is aimed at traffic parameter detection by means of digital image processing technology.Traffic parameter detection which based on image recognition technology acquires vehicle model, vehicle speed, vehicle license plate etc. Many parameters are very important in wide application of Intelligent Transport Systems. In this paper, we only study vehicle license plate recognition and vehicle model detection of traffic information detection system.In vehicle license plate recognition aspect, the position of vehicle license plate is obtained by means of edge detection and characteristics of scan line. Moreover, we introduce Da-Jin method, Radon transformation, erode and dilate operation to obtain the precise position of vehicle license plate and use vertical projection to segment license plate characters. We choose the feature of rough grid as the feature of characters recognition, and directly input the improved unified characters primitive feature to BP neural network classifier to recognize the license plate characters.In vehicle model detection aspect, a new vehicle detection system is designed using algorithms for fast adaptive background extraction, update and background subtraction with contour track to improve the precision of vehicle localization. The vehicle characteristic data are extracted by using binary algorithm, edge detection algorithm, erode and dilate operation, and then the vehicles can be classified by means of D-S evidence theory. Experimentation shows that the method is reliable and efficient.In this paper, we mainly discuss the structures, functions, design and implementation of vehicle license plate and vehicle model recognition system.更多还原