《计算机应用论文基于视频的车型识别中关键技术的应用研究.doc》由会员分享，可在线阅读，更多相关《计算机应用论文基于视频的车型识别中关键技术的应用研究.doc（3页珍藏版）》请在三一文库上搜索。
1、基于视频的车型识别中关键技术的应用研究 基于视频的车型识别中关键技术的应用研究 【中文摘要】 智能交通系统已经成为电子信息技术在交通运输领域的热点研究课题。而图像处理和模式识别技术在智能交通系统的应用研究,是智能交通系统重要的前沿研究领域,具有极其重要的理论和应用价值。随着视频设备在交通系统中的大量普及,结合视频设备,获取更多的交通数据,提高设备利用率成为当前国内外研究的一个重要课题。同时,也是为了克服传统的运动车辆检测和人工识别车辆的诸多缺点的影响,如效率低下、准确度不高、维修安装困难等,所以开展了本课题的研究。基于视频的车型识别系统利用数字图像处理与计算机视觉相结合的方法,获得智能交通系统
2、所需要的信息。它是对特定地点和时间段内的车流信息进行采集、识别和分类,把得到的交通流数据作为交通管理、收费、调度、统计的依据。本文基于图像处理和模式识别的研究现状,在深入研究了已有的运动目标检测算法和模式识别方法及相关理论的基础上,以普通监控用摄像头同步拍摄的运动车辆图像序列为研究对象,对基于视频图像的车辆识别系统中的关键技术-运动车辆检测和目标车辆的识别进行了全面地介绍,着重研究了运动目标检测算法、图像分割算法、识别特征提取和识别方法等。本文分析了不同拍摄角度拍摄到的视频图像对检测和识别的影响,介绍了图像处理中最基本的操作,膨胀、腐蚀、开、闭等操作,重点论述了连通域标记方法,并给出了一种基于
3、轮廓跟踪的连通域标记算法。在视频图像获取过程中,摄像机轻微抖动、天气引起的光线变化等外界环境会影响运动目标的检测精度,针对这一问题,本文提出了一种基于高斯平均的背景更新算法,能较好地抑制外界环境变化带来的影响,并具有较好的实时性。文中采用了基于虚拟检测线框的运动车辆检测算法,并对传统的该算法做了改进。车型识别部分,本文以车辆的正侧面为研究对象,以轮廓的整体形状作为识别特征,提出了一种是基于轮廓形状特征的车型分类识别方法,该算法通过长宽比、拉长度、矩形度、圆形度等车辆轮廓形态的七个特征分量构成一个特征向量组,再以最近邻法作为依据实现车型的识别分类,取得了较为理想的效果。本文以VC+6.0为平台,
4、编程实现了基于视频的车型识别系统,通过实验数据分析表明,本文给出的识别方法能得到较好的识别结果【英文摘要】 At present,intelligent transportation system(ITS) has become a hot research task which is applied to the electronics in the traffic field in the world.Along with the study on application of ITS, image processing and pattern recognition stand on t
5、he front research fields of ITS,and have great theory and application value.With the video equipment greatly used in the traffic system,how to get more traffic data and improve the equipment capacity factor have become an important topic in the world.And more,for overcoming the shortcomings of the t
6、raditional detection and classification,such as the difficulty of installation and maintenance,subjection factors in the manual identification,this paper mainly study on this topic.Vehicle Classification based on video sequences image obtain the information of ITS by using the digital image processi
7、ng and computer vision.It collects,identifies and classifies the vehicles at the specific location and time,and takes it as the traffic management,fees,scheduling and statistics.Based on the research status of the image processing and pattern recognition,this thesis deeply analyzes the previous algo
8、rithms of objects detection and pattern matching,take the sequence of video image by the ordinary video monitoring as the main research objects,and describe the two key technics-vehicle detection and identification,especially study the algorithms of object detection, image segmentation,feature extra
9、ction and recognition method.Then,this thesis analyzed the results with the images taken in different angles,and introduced into many basic operations,such as dilation,erosion etc.But,this thesis takes more attention on the connected component labeling and gives an algorithm,connected component labe
10、ling based on contour tracking.In the process of video obtaining,the variation of outside environment including little dithering of vidicon and weather changing will affect the precision of moving target detection.Aiming at these problems,this paper presents a method of background reconstruction,in
11、which there are moving targets in scene.The background can be dynamically changed.It can decrease the influence of outside environment variation and shorten the time of measurement.In addition,this paper modified the traditional approach,moving vehicle detection based on virtual wirefram.The vehicle
12、 classification takes the features extracted from the vehicle lateral image,and takes the whole contour shape as the recognized features.This paper gives an approach,which is vehicle classification based on characteristics of contour.Eigenvector has been first structured by using length to width rat
13、io,posture ratio,rectangle degree,elongation degree and roundness degree.Then the vehicle type recognition and classification is successfully realized by nearest neighbor method.The technique mentioned above is implemented by experiments with VC+6.0.The experimental results show that the technique h
14、as the advantages of high precision and less calculation.【中文关键词】 智能交通; 运动检测; 车型识别; 连通域标记; 特征提取 【英文关键词】 Intelligent Transportation; Moving Detection; Vehicle Recognition; Connected Component Labeling; Feature Extraction 摘要 2-3 ABSTRACT 3 引言 7-9 第一章 绪论 9-15 1.1 基于视频的车型识别系统简介 9-11 1.2 国内外研究现状 11-14 1.3
15、 论文结构 14-15 第二章 视频序列中的图像处理 15-26 2.1 视频图像的采集 15-16 2.1.1 视频采集常用设备 15 2.1.2 视频序列图像成像原理 15-16 2.2 图像简介 16-19 2.2.1 图像的颜色空间 16-19 2.3 视频图像的随机噪声 19-20 2.3.1 中值滤波算法 19-20 2.4 形态学处理 20-25 2.4.1 数学形态学概述 20-21 2.4.2 形态学基本操作 21-22 2.4.3 连通域的划分 22-25 2.5 小结 25-26 第三章 基于视频的运动车辆检测 26-32 3.1 运动目标检测算法 26-28 3.1.1
16、 背景差分法 27-28 3.2 基于虚拟线框的运动检测算法 28-31 3.2.1 原理 28 3.2.2 检测区域的确定 28-29 3.2.3 算法流程 29-31 3.3 小结 31-32 第四章 基于形状特征的车型识别方法 32-46 4.1 概述 32-33 4.1.1 图像模式识别 32-33 4.1.2 图像的特征提取 33 4.2 特征提取 33-36 4.2.1 边缘检测算法 33-34 4.2.2 特征描述 34-36 4.3 基于形状的目标特征的提取 36-39 4.3.1 矩特征的提取 37-38 4.3.2 形状统计特征 38-39 4.3.3 模板样本库的建立 39 4.4 分类识别算法 39-45 4.4.1 识别算法简介 40 4.4.2 基于近邻法的车型识别 40-42 4.4.3 实验过程 42-44 4.4.4 实验结果及分析 44-45 4.5 小结 45-46 第五章 总结与展望 46-48 5.1 总结 46-47 5.2 展望 47-48 参考文献 48-51 攻读学位期间的研究成果 51-52 致谢 52-54