基于交通视频的车辆检测与追踪算法研究

发布时间：2018-05-31 00:33

本文选题：智能交通系统 + 背景更新阴影与光斑去除拥堵检测　；参考：《华中科技大学》2016年硕士论文

【摘要】：随着时代进步,智能交通系统(ITS)为解决交通问题发挥了重要作用。实时的车辆检测与追踪是ITS必不可少的环节,也是研究热点之一。本文主要对高速公路下的视频流进行采样分析,详细描述了该场景下的车辆检测与追踪算法的实现。本文对比分析了已有的车辆检测与追踪技术,对相关算法做出了适当的改进并给出了改进后的实验结果。下面这几个方面是本文的主要工作内容:(1)背景更新算法的研究由于系统实时性的要求,本文采取的背景差分方式来实现运动目标车辆的检测。本文介绍了常用的背景建模方法,对比分析了其中的优缺点,并设计了一种分区块的自适应背景更新算法。该算法利用的是加权累计帧间差方式来实现背景的自动更新。该算法的优点在于计算量相对较小,背景更新速率较快,初始化建立背景的时间相对较短。(2)日间阴影以及夜晚光斑的去除汽车白天行驶期间,由于日光的照射会在地面形成影子,这些运动的影子对准确的提取运动车辆轮廓产生了极大的干扰作用。汽车在夜间行驶,由于近光灯的照射在地面形成明亮的光斑,这些光斑也会对检测结果造成恶劣的影响。本系统有效解决了阴影和光斑所带来的干扰,同时还解决了系统自动区分白天与夜间时间段的问题(3)拥堵与非拥堵状态下的车辆检测本文在道路拥堵状态下利用虚拟线圈的技术来检测运动车辆,非拥堵状态下则采用基于检测线的技术。本文对比了虚拟线圈和检测线技术的优缺点和不足,指出了各自适合的应用场景。本文提出了一种简易的方法来实现道路拥堵与非拥堵状态下检测算法的自动切换。(4)车辆追踪算法的研究本文在以往的车辆追踪技术上,设计了一种基于车辆特征的追踪算法。由于在经过二值化处理后,已经提取出运动车辆的目标轮廓,因此光照的影响已经可以忽略不计,从而利用车辆的轮廓特征实现目标追踪。本文解决了在追踪过程中由于运动目标车辆重叠,导致追踪失败的问题。由于从3D投影至2D平面导致信息丢失,图像中的车辆像素距离不等同于真实距离。因此对车速和位置的估算仍然缺乏准确性,需要进一步的深入研究。在以上的研究工作基础上,本文设计了完整的基于交通视频的车辆检测与追踪系统,给出了实验结果。结果表明,该系统在车辆检测上具有很好的实时性与准确性,实现了初步的跟踪效果,取得了令人满意的成绩。
[Abstract]:With the progress of the times, Intelligent Transportation system (its) plays an important role in solving traffic problems. Real-time vehicle detection and tracking is an essential part of ITS, and also one of the research hotspots. In this paper, the video stream under the freeway is sampled and analyzed, and the realization of the vehicle detection and tracking algorithm in this scene is described in detail. In this paper, the existing vehicle detection and tracking techniques are compared and analyzed, and the related algorithms are improved and the experimental results are given. The following several aspects are the main work of this paper: 1) background updating algorithm; because of the real-time requirement of the system, the background differential method is adopted to realize the vehicle detection of moving targets. In this paper, common background modeling methods are introduced, their advantages and disadvantages are compared and analyzed, and an adaptive background updating algorithm is designed. The algorithm uses the weighted cumulative frame difference to realize the automatic updating of the background. The advantages of the algorithm are that the computation is relatively small, the background updating rate is faster, the time of setting up the background is relatively short) the daytime shadow and the removal of the night spot during the day driving of the vehicle. Since sunlight can form shadows on the ground, these moving shadows have a great interference effect on accurate extraction of moving vehicle contours. At night, because of the bright spot formed on the ground by the near-light, these spots will also have a bad effect on the detection results. The system effectively solves the interference caused by shadow and spot. At the same time, it solves the problem that the system can automatically distinguish the day and night time periods. (3) vehicle detection in congested and non-congested state. In this paper, virtual coil technology is used to detect moving vehicles under road congestion. In non-congestion condition, the technology based on detection line is adopted. In this paper, the advantages and disadvantages of virtual coil and detection line are compared, and the suitable application scenarios are pointed out. In this paper, a simple method is proposed to realize the automatic toggle. 4) vehicle tracking algorithm for road congestion and non-congestion detection algorithms. In this paper, a vehicle tracking algorithm based on vehicle characteristics is designed in the previous vehicle tracking technology. Since the target contour of moving vehicle has been extracted after binarization, the influence of illumination can be neglected, so the contour feature of vehicle can be used to track the target. In this paper, the problem of tracking failure caused by the overlap of moving target vehicles in the tracking process is solved. Due to the loss of information from 3D projection to 2D plane, the distance of vehicle pixels in the image is not equal to the real distance. Therefore, the estimation of speed and location still lacks accuracy and needs further study. Based on the above research work, a complete vehicle detection and tracking system based on traffic video is designed, and the experimental results are given. The results show that the system has good real-time and accuracy in vehicle detection, and has achieved a preliminary tracking effect and achieved satisfactory results.
【学位授予单位】：华中科技大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：U495;TP391.41

【相似文献】