【作者】 汪泉；

【导师】 王命延；

【作者基本信息】 南昌大学 ， 计算机应用技术， 2007， 硕士

【摘要】 随着计算机硬件技术和计算机视觉技术的发展，基于实时图像处理的交通监控系统成为当下的发展趋势，而视频车辆的检测与跟踪是智能交通系统的核心部分。本文研究的是利用图像处理技术对高速公路上的车辆进行实时监控，采用基于运动车辆检测和跟踪的方法来对车辆进行测速，为智能交通系统提供交通参数。本文对摄像机标定方法做了研究，并采用一种标定方法，并提出通过检测车道线中的一些点来完成坐标输入。本文提出了针对运动车辆进行实时检测和跟踪的算法，适用于大面积、多目标的复杂场景，能排除干扰，主要应用于高速公路上。本文提出了在车辆检测之前先做行车区域检测，通过行车区域检测排除行车区域外的干扰。然后通过实验数据对于目前几种有代表性的背景模型进行了比较，找出其中的优缺点，结合这些方法进行改进后，提出了基于亮度与亮度梯度信息的混合高斯模型。该背景模型较好的解决了背景模型的初始化、更新、背景干扰、外界光照等问题，在背景、前景的判断上，充分利用了亮度与亮度梯度信息。使得系统的背景模型既能够满足背景随时间渐变的统计特性，又能够兼顾系统的噪声以及一些突发的干扰因素。在背景差方法上提出了基于三层结构的背景差算法，从三个层次来保证车辆检测的准确性。在运动目标跟踪方面，本文使用了基于扩展卡尔曼滤波器的目标跟踪，并对滤波器的构造、更新做了改进，然后提出了三种匹配原则，综合预测和搜索匹配，并考虑遮挡问题提出了一个完整的跟踪算法。最终根据跟踪的结果可以计算出车辆的速度。本文最后还提出了视频测速系统的设计方案。总之，本文对视频测速目标检测和跟踪问题进行了深入的分析和研究，提出了自己的方法去解决这些问题，并且对多个方法作了实验，实验结果表明，本文提出的方法能够满足实际系统的的要求。更多还原

【Abstract】 With the development of computer hardware and computer vision technologies, the traffic surveillance system based on the real-time image processing becomes the immediately development tendency,and the detection and tracking are the important part of the Intelligent Transport System.This article studied that used the image processing technology to carry on the real-time monitoring to the highway vehicles, used the method based on moving object detection and tracking to calculate the speed of vehicles, so provided the transportation parameter for the Intelligent Transport System.This article researched the methods of Camera Calibration,and adopted one method, and proposed we input coordinates through the detection of some point of lane edge.This article proposed we carried on the real-time detection and tracking algorithm in view of the moving vehicles, that was suitable to the big area, the multi-goals complex scene, and could remove the disturbance,and the algorithm mainly applied on the highway.This article proposed that the system made firstly the detection of driving region before vehicles detect, and the system could remove the disturbance outside the driving region through the detection of driving region. Then we had the present several kinds representative background model through the empirical datum to carry on the comparison,discovered good and bad points, after unified these methods to make the improvement, proposed a improved Gaussian mixture background model based on intensity and the gradient of intensity. This background model better soluted the problems such as model initialization, update, background disturbance, illumination, and it had fully used intensity and the gradient to distinguish the background and foreground.The background model both could satisfy the statistical property which the background changed gradually as necessary, and could remove the disturbances of noise of system and some sudden factors. The article proposed an algorithms of background difference based on three-level structure,and the algorithms guaranteed the accuracy of the detection of vehicles from three levels.In the tracking of moving object, this article used the method based on EKF, and improved the structure and update of the filter, then proposed three kinds of match principles, the article synthetized the forecast and the match, and proposed a complete track algorithm under occlusion. Finally the system could calculate the speed of the vehicles according to the result which tracks. A vehicle speed measurement system through video were designed in this article.The article analysed the problem of distance measuring and moving objects detecting and tracking, and puts forward some way to solve them and test them through experiments. The result from the experiments gave a positive answer that these methods met real-time and practicality requirements.更多还原