【作者】 何静；

【导师】 汪国有；

【作者基本信息】 华中科技大学 ， 模式识别与智能系统， 2011， 硕士

【摘要】 交通视频监控系统通过从交通监控视频中捕获目标、提取目标信息并进行分析处理达到对城市交通的监控管理。本文主要对智能交通监控系统的架构进行了研究,对系统中涉及的图像处理关键技术如车辆检测算法、车辆测速算法以及车牌识别算法等进行了相关研究,并予以实现。在系统架构方面,本文提出了一种基于车辆基本信息元的系统架构。这种系统架构可以从本质上体现监控系统管理的基本目标为车辆目标,将整个系统从车辆检测到车牌识别的流程统一为对车辆信息元属性特征的提取管理。不仅为系统提供了更好的可扩展性,同时也为系统的管理提供了很大的方便。在车辆检测算法的研究和实现过程中,本文主要实现了一种基于背景差和帧间差的运动区域检测算法,并针对道路交通中易出现的车辆粘连情况,实现了一种基于凹点检测的粘连分割方法,能较好的处理车辆检测中常遇到的阴影和粘连问题。在车辆测速算法的研究中,本文提出一种基于道路标线检测的车辆测速算法,该算法依据世界坐标系到图像坐标系的转换关系,省略了摄像机标定的繁复过程,并根据车辆信息元的位置属性,可以计算图像中任意两点之间的距离,从而大大提高了车辆测速的精度。在车牌识别算法方面,将车牌作为车辆信息元的一个基本属性,本文实现了基于增强边缘的形态学车牌定位算法和基于Radon变换的车牌倾斜校正算法。在车牌字符分割算法的研究过程中,提出了一种基于灰度重心的模板分割方法。参考车牌字符中数字字母的应用特征,实现了基于多级模板匹配的数字字母分类器,在试验中取得了较好的效果。更多还原

【Abstract】 Intelligent Transport Registration System is a process to automate retrival and analysis vehicle information from videos taken at crossroads or highway, in order to strengthen the control of road traffic, public transport. In this thesis, we focus on the system architecture of ITS, and the relate algorithm realizing.We proposed a system architecture based on vehicle informationunit. The architecture embody that the vehicle is the transport registration system’s essence. This system can keep balance among the whole procedure, to achieve a comprehensive effect. It also strengthens the extensibility, reconfiguration and portability of the system. provide facilities with respect to the management of system.In the research process of vehicle detect, we realize a detect algorithm based on background subtraction and frame difference. In the light of vehicle adherent, a algorithm based on concave spot detecting is implemented, the results shows that it is a good way to deal with the problem. We proposed a method of vehicle velocity measure based on road traffic marking detecting, we take advantage of the transformational relation of World Coordinate and Image Coordinate, omit the step of camera calibration. Not only enhance the efficiency of system, but also enhance the accuracy of velocity measure by averaging the outcome of repeated measuring.Besides, we take the plate number as a base attribute of vehicle informationunit. This thesis take a comparison among traditional plate number allocation method, realize a method of locate based on strengthened edge determination, a method of angularity correction based on Radon transform. In order to get the separate char from the string, we proposed a char segmentation method based on the template of gray-center. Refer to the property of numberalphabet and chinese character in plate, we realize a numberalphabet classifier based on multilevel template matching .更多还原