【作者】 俞亮；

【导师】 尚群立； 孔亚广；

【作者基本信息】 杭州电子科技大学 ， 控制理论与控制工程， 2010， 硕士

【摘要】 随着世界各国经济的快速发展,车辆的数量急剧增加,由此引发的交通状况问题日益受到人们的重视。智能交通系统就是为了解决这个问题而诞生的,目前已经成为全球各国政府和有关部门高度重视的高科技新领域。基于视频的闯红灯车辆自动检测系统是智能交通系统中的重要组成部分,其中有效的检测和实时跟踪是车辆的行为分析和行为判断的前提。鉴于以上背景,本文对闯红灯车辆检测与跟踪算法进行研究。在分析和总结现有算法的基础上,提出了两个改进算法。本文针对传统的背景提取方法在前景运动对象密度较高,亦或前景出现过于频繁的情况下,提取到的背景图像中会掺杂很多前景成分的问题,提出了一种基于形态学和像素灰度归类的背景重建算法。该算法通过三帧差分法和形态学把每帧图像区分为前景对象和背景区域。然后在用像素灰度归类方背景重建时不考虑前景对象,提高了背景出现的频率,从而能正确重建背景。通过实验证明,新方法能够在车辆密度较大的情况下能正确的生成背景。本文为了满足车辆跟踪实时的要求,设计了一种自动初始化窗口的Mean Shift算法。另外针对Mean Shift算法在目标的颜色分布和背景相似情况下,会丢失目标的问题,利用kalman滤波器根据前面的目标位置信息来预测在本帧图像中目标的可能位置,然后用Mean Shift算法在这个位置的邻域内找到目标的真实位置。这样,我们利用了Mean Shift所没有利用的前面一帧目标的运动信息,丰富了对已知信息的使用,增强了跟踪效果。通过实验证明此算法是有效性。更多还原

【Abstract】 With the rapid development of the world economy, the number of vehicles increase dramatically to lead to the traffic problem that is attented. Intelligent transportation system is born for solving this problem and now the global governments and relevant departments pay close attention to this new high-tech area. Automatic detection system of red light running vehicles with Video is an important component of the intelligent transportation system, in which an effective detection and real-time tracking are determine premise of behavior analysis and behavior judgment in vehicles.View as the above background, this paper studies detection and tracking of red light running vehicle algorithm. In basis of anglicizing and summarizing existing algorithms, it proposes two improved algorithms.Against the problem that the tradition background extraction method could not precise extraction background when moving objects have high density or appear too frequently. In this paper presented an algorithm of background reconstruction based on morphological and PIC (Pixel intensity classification). The algorithm classify the pixels in each frame into background area and moving objection area using three frame difference and Morphological analysis. Then extracting background modeling with PIC except the moving objection area, raised the frequency of background, so the background could be rebuilded proper. The experiment results show that the method proposed in this paper could extract accurate background in high vehicle density.In order to meet the requirements of real-time vehicle tracking, it designs an Mean Shift algorithm of automatic initialization window. Another in the cases of the similar of target color distribution and background, it considers the Mean Shift Algorithm will lose target. So, according to position information in front, kalman filter is used to predict the possible position of target image in the frame, and then at the location of the neighborhood you can use the Mean Shift to find the true target location. It enriches the use of known information and enhances the tracking results by the use of target movement information of before frame. Experiment verifies the effectiveness of the method.更多还原