【作者】 张硕；

【导师】 杨耀权；

【作者基本信息】 华北电力大学 ， 模式识别与智能系统， 2014， 硕士

【摘要】 随着社会的快速发展和城市的不断扩张，人口增多的同时，车辆总数也在不断增加中，造成了道路交通系统压力不断的增大。因此对智能交通系统(Intelligent Transportation System, ITS)研究的重要性和必要性日益突出，系统的核心是基于视频的车流量检测与车辆跟踪的研究。本文主要研究目标是基于机器视觉技术，对道路上行驶的车辆进行检测、识别、跟踪以及车流量统计。本课题对运动车流量的检测与车辆的跟踪进行了研究，分析和总结了现有的检测与跟踪技术，并在国内外相关研究成果的基础上，对相应的算法进行优化和改进。具体的研究工作有：首先，对采集到的视频图像进行预处理，采用改进的混合高斯背景更新算法，将当前帧图像与背景图像逐像素点进行相减即得到前景图像(即运动目标图像)，并对前景图像采用最大类间方差法进行阈值分割，得到二值图像，从而利用矩形框提取运动目标车辆；然后利用基于运动目标质心的特征匹配与Mean-shift迭代相结合的算法对提取到的目标进行跟踪；同时对每一个检测到的目标跟踪链进行计数，完成车流量统计。软件上，使用Visual Studio2005开发平台，基于MFC框架下，利用Open CV的数据结构、函数，对车流量检测与车辆跟踪系统进行设计与实现。在本文算法的基础上，对两种不同角度的视频进行了检测与跟踪。实验研究结果表明，车流量检测结果准确率平均水平保持在90%以上，且跟踪快速实时性强，能够达到预期的效果，验证了算法的稳定性。更多还原

【Abstract】 With the social development and urban expansion, population increase, whilealso increasing the total number of vehicles, resulting in a traffic system pressurecontinuously increases. Therefore, the Intelligent Transportation System (IntelligentTransportation System, ITS) highlights the growing importance of research, its coreis video-based vehicle flow detection and vehicle tracking study. The main goal ofthis paper is based on machine vision technology to detect, identify, track the roadvehicles, and get vehicle flow statistics.The issue of moving traffic flow detection and vehicle tracking were studied,the existing detection and tracking technology were analyzed and summarized, andat the same time, based on research achievements at home and abroad, thecorresponding algorithms are optimized and improved. The specific studies are: thecollected video image is preprocessed, and using an improved background updatingalgorithm of Gaussian Mixture Model, that the current frame image subtracts thebackground image can obtain the foreground image(i.e. moving target image), thenprocess it using Otsu threshold segmentation method to extract the moving targetvehicle; then combine centroid-based feature matching and Mean-shift iterativealgorithm to track the extracted target; at the same time, count for each detectedtarget tracking chain to complete traffic flow statistics. In software, at the VisualStudio2005development platform, based on the MFC framework, using Open CVdata structures, functions, designs and achieves the system about vehicle flowdetection and vehicle tracking.On the basis of this algorithm, detect and track the two different angles ofvideo. This algorithm experimental results show that, the accuracy of vehicle flowdetection remained above ninety percent, with timeliness strong to track, and thesystem achieved the desired results, verified the stability of the algorithm.更多还原