基于计算机视觉的车辆跟踪算法研究

发布时间：2018-06-05 18:18

  本文选题：动态背景建模 + 卡尔曼滤波　； 参考：《广东工业大学》2014年硕士论文

【摘要】：目标跟踪技术被提出来已经有几十年的时间,经过长时间的研究与发展,它已经成为当今社会非常重要的高端技术,它在日常生活和军事应用领域都发挥着巨大的作用。此外,伴随着计算机视觉技术的不断发展,基于计算机视觉的车辆检测与跟踪技术成为智能交通系统的研究热点,是视频监控中动态目标的流量统计、行为分析的重要理论基础。对于动态背景环境下多目标的检测与跟踪,仍旧是一个开放的研究领域,很多科研机构和高等院校都对这一领域的研究投入了大量的时间和精力。本文主要研究了交通视频监控系统中运动目标的检测与跟踪技术,对前景分割、车辆检测、卡尔曼滤波、多目标数据关联等问题进行了比较深入地研究。这一技术在智能交通监控系统和智能视频监控系统中有着广泛的应用前景,具有极高的研究价值。 本文主要研究内容包括以下几个方面： 1、运动车辆的检测：首先介绍了几种常用的前景分割算法,本文采用基于简化KDE(Kernel Density Estimation)的动态背景建模方法,该方法能够在复杂的背景环境下有效地提取出运动前景,通过去噪和连通域处理,使得运动车辆的检测十分准确和有效,为接下来的目标跟踪做准备。 2、运动车辆的跟踪：在卡尔曼滤波算法的框架下进行单目标跟踪的相关研究,在特征选择部分,提出了一种rgI颜色直方图,通过实验对比说明它相对于HSV(Hue,Saturation and Value)空间矩特征具有更好的跟踪定位效果,能够完成复杂背景环境下的目标跟踪过程。通过实验比较不同特征的跟踪效果,验证了rgI颜色直方图的优越性,它能够作为一种非常有效的特征对运动目标进行描述。 3、多目标数据关联：本文通过对检测到的多目标提取全局特征并进行特征匹配关联,建立相似函数,对多个测量值进行寻优判断,求得概率最大的测量值,并通过与相关阈值进行比较来作为匹配结果,从而完成多目标的跟踪过程。 大量实验结果显示,本文提出的跟踪算法实时性高,鲁棒性强,能够准确处理真实交通场景环境下多车辆跟踪,具有很高的应用价值。
[Abstract]:Target tracking technology has been put forward for several decades. After a long time of research and development, it has become a very important high-end technology in today's society, it plays a great role in daily life and military applications. In addition, with the continuous development of computer vision technology, vehicle detection and tracking technology based on computer vision has become the research hotspot of intelligent transportation system, which is the traffic statistics of dynamic target in video surveillance. The important theoretical basis of behavior analysis. Detection and tracking of multi-targets in dynamic background environment is still an open research field. Many research institutions and universities have invested a lot of time and energy in this field. This paper mainly studies the technology of moving target detection and tracking in the traffic video surveillance system, and deeply studies the problems of foreground segmentation, vehicle detection, Kalman filter, multi-target data association and so on. This technology has a wide application prospect in intelligent traffic monitoring system and intelligent video surveillance system, and has high research value. The main contents of this paper include the following aspects: 1. Detection of moving vehicles: firstly, several common foreground segmentation algorithms are introduced. In this paper, a dynamic background modeling method based on simplified KDE(Kernel Density estimation is used, which can extract the motion foreground effectively in complex background environment. By denoising and connected domain processing, the detection of moving vehicles is very accurate and effective. 2. Tracking of moving vehicles: the research of single target tracking is carried out under the framework of Kalman filter algorithm. In the part of feature selection, a rgI color histogram is proposed. The experimental results show that it has a better tracking and localization effect than the spatial moment feature of HSV HueSaturation and, and it can complete the target tracking process in complex background environment. The advantages of rgI color histogram are verified by comparing the tracking effects of different features. It can be used as a very effective feature to describe moving targets. 3, multi-objective data association: by extracting the global feature and matching the feature to the detected multi-target, the similar function is established, and the most probabilistic measurement value is obtained. The multi-target tracking process is completed by comparing with the correlation threshold as the matching result. A large number of experimental results show that the proposed tracking algorithm is highly real-time robust and can accurately handle multi-vehicle tracking in real traffic scene environment. It has high application value.
【学位授予单位】：广东工业大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP391.41;U495
，

本文编号：1982975