【作者】 李卫江；

【导师】 郭晓汾；

【作者基本信息】 长安大学 ， 载运工具运用工程， 2008， 博士

【摘要】 智能交通系统ITS(Intelligent Transportation System)是新一代的交通信息管理和控制系统，它充分利用现代电子、控制、计算机和数据通讯等技术，能大大提高交通运输的安全性和运行效率，及时提供道路车辆流量和路况信息，记录违章车辆，增强突发交通事故的处理能力，为人们的出行提供快捷、舒适的交通服务，从而促进交通运输事业的迅速发展。作为智能交通系统的重要组成部分，道路交通信息的实时采集是智能交通信息采集的关键内容之一。及时、准确地获取各种交通参数是实现交通控制智能化的必要前提条件，道路交通检测器及其检测技术的高低将直接影响道路交通系统的整体运行效率和管理水平。由于基于图像分析及计算机视觉为基础全天候对车辆进行监测和识别的技术逐渐成熟，交通检测也由人工观察交通现象、手动控制交通信号转向视频检测。视频检测方法能够利用视频跟踪车辆的手段，自动检测以交通流量为主的交通参数，实现交通信号控制并对危险事件进行报警，无疑它是一种有发展潜力的交通信息采集方法。本论文对基于线阵CCD(Charge-Coupled Devices)成像检测系统的几个关键技术进行了研究，着重就交通场景下的背景提取、交通视频图像中车辆的检测和跟踪、车牌识别以及交通视频监控系统的实现等重要而基本的问题进行了深入的探讨。具体说来，本论文所做的主要研究工作可概括如下：(1)提出了一种基于小波变换的背景提取方法及利用平均方差代替均值的Otsu阈值选取方法。结合小波变换的特点，针对线阵CCD扫描图像，提出了一种基于小波变换的背景提取方法。该方法能够有效的消除灯光等噪声，能够较好的提取出背景信息。同时，针对在背景提取过程中，阈值的选取对背景的提取具有较大的影响，本论文利用图像的灰度平均方差反映了图像灰度分布的特性，提出了一种利用平均方差代替均值的Otsu阈值选取方法，即改进的Otsu分割方法。(2)提出了一种基于主动轮廓外力场模型PMF(Perona-Malik Field)的车辆分割方法。该方法利用经典的Perona-Malik各向异性去噪模型具有保护边界信息的特点，将经过Perona-Malik模型处理后图像的负梯度作为外力场，研究其对车辆分割结果的影响，从而提出了一种基于主动轮廓外力场模型PMF的车辆分割方法。该方法不仅能够保持车辆的边界信息，克服了传统外力场不能进入车辆图像凹部的缺陷，而且对初始曲线的约束较少。(3)提出了一种加权迭代车辆匹配方法。现有的大部分匹配方法都是基于均方误差最小的原理，即最小二乘法。当噪声是高斯噪声时，最小二乘法是最优的，但在实际过程中，经常会出现一些出格数据，若把出格数据当作高斯噪声，那就有可能导致错误的结果。针对这种噪声，本论文提出了一种加权迭代车辆匹配方法，该方法首先利用加权来进行车辆匹配，再利用各点余差的倒数作为下一次迭代的权值，如此循环，就可以使出格数据的权值接近于0，最后完成高精度的车辆匹配。本方法可以克服最小二乘法鲁棒性差及随机抽样方法计算量大的缺点，模拟实验和真实实验数据结果表明，该匹配方法具有运算量小、鲁棒性好等优点。(4)提出了一种用于牌照识别的图像增强方法。牌照区域分割是牌照识别的关键步骤，增强图像中的牌照区域，抑制背景区域，可以有效降低牌照区域分割的难度。本方法将图像分解为一组二值图像的组合，然后在二值图像上计算各连通分量及其特征参数，利用牌照区域和背景区域对应的连通分量的特征差别，可以有效抑制背景而保留牌照。处理后的二值图像可重构出牌照区域增强的部分。本论文还采用等高线标记代替连通分量标记，以减少计算量，使得该方法具有实用性。实验显示，该方法能够有效地突出了牌照区域而抑制了背景，提高了牌照定位分割的效果，可以很好地用于实际的牌照识别系统中。(5)实现了一种基于线阵CCD摄像机的交通信息采集和检测系统。交通信息是交通管理和规划的基础和关键，实时动态地采集描述交通流的各种参数和检测各种交通事件，并对这些信息进行有效地处理和应用，对提高交通系统的运行效率、减少交通事故、提高紧急事件的应急能力、优化交通系统的规划设计和有效评价交通系统的运行指标具有很重要的作用。为了获取交通信息，本论文实现了一种基于线阵CCD摄像机的交通信息采集和检测系统。该系统主要由数据库、图像库、视频触发、图像抓拍、断面计时、参数获取和牌照识别等主要功能模块组成，具有视频触发、瞬时速度检测、超速实时处理、车型识别和交通量统计、车辆通行记录等功能。更多还原

【Abstract】 Adopting advanced techniques in electronics,automation,computer and digitalcommunication,etc.,the Intelligent Transportation System,i.e.ITS,represents the novelgeneration of traffic information managing and controlling system,which can provide trafficinformation of volume of vehicles or state of roads in time,keep records of traffic irregularity,enhance ability of sudden traffic accidents,promote traffic safety & efficiency together withconvenient services for public travels and better improvements of environmental quality andtherefore advances national traffic transportation cause greatly.As one of essential parts in ITS,the real-time collection of road traffic information playsa vital role in ITS information collection.It is a necessary premise to obtain various trafficparameters timely and precisely and level of road traffic detectors and correspondingdetecting techniques have direct effect on management or running level of the whole roadtraffic system.With the maturity of all-sided vehicle surveillance and detection techniquesbased on image analysis and computer vision,development of traffic detection was alsochanged from observation of traffic phenomena and traffic signal controlling manually towhich based on video detection.Vehicle detection based on video method could detect themain traffic parameter of traffic volume completely by means of video vehicle tracking,control traffic signal automatically,and alarm with dangerous occurrence.It is undoubtedly amost lively traffic information collection method with deep potential.This thesis has performed some researches on several vital techniques of trafficparameters detecting system based on line-CCD camera.It emphasizes on detailed discussionabout the important and basic issues under traffic scenes such as background extraction,videovehicle detection and tracking,plate recognition,realization of traffic surveillance system anddata stream simulation etc.For these problems,some deep viewpoints or actual realizationmeans is presented and necessary test and simulation is performed.In detail,the main research work finished in the thesis can be summarized as follows:1.A background extraction method based on wavelet transformation and an improvedmethod for Otsu threshold are presented.Combining the feature of wavelet transformation,abackground extraction method is presented based on image from line-CCD camera.Theexperimental results show that the proposed method can effectively polish the noise from lamp and extract the background.At the same time,background extracted is depended on thethreshold election.And the average of image grey represents the distributing of it.So,theimproved Otsu threshold method that the average is instead of variance is presented.The realexperimental results show that the improved Otsu threshold method can effectively extract thebackground.2.An external force field for active contour model-PMF is presented.ThePerona-Malik model which is a classical method of anisotropic removing noises has theadvantage of remaining the edge map of image.The negative gradient of restored image byPerona-Malik model is defined as external forces,and the segmentation results that it affectson active contour model are studied.Accordingly,an external force field for active contourmodel-PMF is presented.Theoretical analyses and experimental results show that PMF canretain the edge information of image and enter into the edge’s concaves entirely.In the meantime,PMF has large capture range with few restrictions to initial curves.Moreover,becausePMF is derived from Perona-Malik,it is robust to the noise.3.A traffic match method based on iterative weight is presented,which can efficientlydiscard the outliers.Now,most of match method is base on least mean-square method.Whenthe noise is gaussian distribution,the solution based on least mean-square method is best.But,when the outliers appear,and the outliers are regard as the gaussian noise,which maybe resultin error solution.So,the outliers should be discarded.The weight of each point is determinedbased on the inverse of error and the error is obtained based on the weight.After severaliterations,the weights of the outliers trend to zero and the match result is obtained with goodaccuracy.The method can overcome the disadvantages of both the least-squares method andthe Random Sample and Consensus method.The theory and experiments with both simulateand real data demonstrate that the method is very efficient and robust.4.A new image enhancement method for car license plate recognition is presented.Segmentation of license plate region is a key procedure in a car license plate recognitionsystem,enhancing the plate region of the captured image and suppressing the backgroundarea can effectively reduce the difficulty of the plate segmentation task.In this paper,theimage used for license recognition is decomposed into a group of binary versions,and thenconnected regions in the binary images are labeled and their feature parameters are calculated.As the connected regions in the plate area and those in the background are very different in the calculated feature parameters,so the connected regions which likely belong to the platearea can be preserved and the others can be erased.A new image with plate region enhancedcan be reconstructed from the processed binary images.In order to reduce the computationload and make the method realizable,contour lines instead of connected component labelingis used in the paper to describe the connected regions.The experimental results show that theproposed method effectively enhances the plate area and suppress the background area,improves the plate segmentation performance.The new method can be well applied in the realcar license plate recognition systems.5.Traffic information is the key and foundation of traffic management and programming.the real-time collection of road traffic information,detection the traffic affairs and processionand application these information and affairs play an important role on improving theefficiency of traffic system,the decrease of traffic accident,increase the ability of solving theemergency affairs and optimization of programming design.To obtain traffic information,atraffic information collection and detection system based on line-CCD camera is presented.The system consists of data and image library,vidio trigger,image capture,sectioncalculagraph,parameter obtaining,plate recognition,etc.And the system can detect the trafficvelocity and flux,recognise plate,recode traffic.At the same time,the method for vehicle andits velocity detection is presented and applied to the system.The experiment results show thatthe system can accurately detect the traffic and its velocity.更多还原