【作者】 范五一；

【导师】 李晋宏； 熊昌镇；

【作者基本信息】 北方工业大学 ， 计算机软件与理论， 2010， 硕士

【摘要】 伴随着图像处理、计算机视觉、模式识别以及人工智能等学科的兴起,基于视频的交通参数检测技术已经成为计算机视觉领域新兴的研究方向,具有较高的学术价值和理论研究意义。本文以路口固定的摄像头所拍摄的视频图像为研究对象,深入研究了图像序列中运动目标检测与跟踪,完成了运动目标的实时检测,并在此基础上采用VC++6.0以及Open CV视觉库图像开发包,开发出了一套完整的视频交通流参数检测系统,取得了良好的实验效果。运动目标检测算法是视频交通流参数检测的主流算法。本文通过分析背景差法、高斯-贝叶斯建模、帧间差法,边缘检测法等运动目标检测算法的技术特点与优势,提出了一种综合交通流参数检测模型。对于标清视频采用高斯贝叶斯混合模型的策略进行交通流参数的检测与跟踪；而对于高清视频(高清视频像素值为2592×1936或者更大),为解决现有的背景建模方法应用高清图像的车辆检测时存在计算量大的问题,提出一种基于亮度曲线的高清图像自适应背景更新算法,该算法首先设置检测区域,然后对检测区域的背景图像和当前帧图像生成亮度曲线,利用两条亮度曲线,检测车辆的存在与否,解决了高清图像在实际应用中背景建模耗时多,不能完成实时检测的难题。实验表明,该算法实时性较高,简单有效,可实时完成高清和标清视频的分车道流量统计,对于50万以内像素的视频可以完成车辆跟踪、车速测定、轨迹生成、逆行以及超车的判断等,通过48小时的现场测试,车流量统计的准确率达92%,车速的测定准确率达80%,车道占有率、是否逆行、行驶方向、超速状态的判定已达到了90%。对于高清视频图像可以完成实时车流量检测,其检测准确率达99.2%。更多还原

【Abstract】 With the advance of image processing、computer vision, pattern recognition、artificial intelligence and other disciplines, traffic parameter detection technology based on video has became the new direction of computer vision field which contains high academic value and significant theory.In this paper, the video images captured by a fixed camera in intersection were used to be objects for this study. The moving target detection and tracking in image sequence was studied in depth and we completed real-time detection of moving targets and on this basis, using VC+ +6.0 and the Open CV library visual image development kit, then we finished the video traffic flow measurement system and obtain good experimental results.Video motion detection algorithm is the mainstream of traffic flow parameters detection algorithm. By analyzing the features and advantages of some moving target detection algorithms such as the background subtraction method、Gauss-Bayesian modeling, frame difference method、edge detection algorithms, we propose a Comprehensive Traffic Flow Detection Model algorithm. Use Gaussian mixture model and Bayesian strategy to do traffic flow parameters of the detection and tracking for standard definition video; while for the high-definition video (pixel value of HD video is 2592×1936, or larger), in order to solve the problem of large computation when using the method of background modeling in the high-resolution images, a new algorithm is proposed which is based on intensity curve of high-resolution image in traffic detection. First of all, set the detection areas, then the intensity curves of the background image and current frame are drawn, which are used to detect the presence of vehicles, so it can solve the problem that time-consuming is enormous and vehicles can not be detected in real-time in high-resolution images.The experiments show that the algorithm is real-time, simple and effective which can accomplish the sub-drive traffic statistics of standard definition video and high-resolution video in real time. This algorithm can accomplish the vehicle tracking、speed determination、trajectory generation、retrograde、overtaking in the video within 500,000 pixels. According to field test of 48 hours, the accuracy of traffic flow statistics reached 92% and the accuracy of speed reached 80%. The accuracy of traffic lane share, retrograde drive detection, driving direction and speeding reached 90%. In the HD video, traffic flow can be detected in real-time and the accuracy is up to 99.2%.更多还原