基于视频的车辆检测与跟踪算法研究

发布时间：2018-06-18 22:18

本文选题：车辆检测与跟踪 + 前景分割　；参考：《北京邮电大学》2016年硕士论文

【摘要】：随着当今社会的不断发展,智能交通系统越来越成为世界各个国家广泛关注的焦点。智能交通系统在道路监控,以及道路交通管理方面上发挥了重要的作用。智能交通系统的其中一个热点和难点是车辆检测与跟踪。论文围绕视频中车辆的运动检测,跟踪以及相关的阴影去除和遮挡处理等重难点问题展开研究,同时将这些方法应用于视频监控的系统中。主要的工作分为以下几点。(1)论文提出改进的背景建模算法以及无参化特征融合阴影去除算法。首先介绍了背景建模算法,选取Vibe(Visual Background extractor)作为系统的目标检测算法,对Vibe建模算法的不足进行了阐述,并提出改进方法。该方法能够有效抑制Vibe算法容易引入的鬼影问题,从而更加准确的对目标进行检测。同时为了消除运动车辆阴影对前景结果造成的影响,提出一种改进HSV(Hue,Saturation,Value)特征与纹理LBP(Local Binary Patterns)特征相融合去除阴影的方法,对HSV特征无参化,此融合算法有效地去除了车辆阴影。(2)论文提出一种基于改进观测模型的粒子滤波跟踪算法。首先研究目标跟踪算法,介绍了几种典型算法,并阐述了算法基本原理,给出对比分析。同时,提出了一种基于改进HOG(Histogram of Oriented Gradients)特征的车辆观测模型,在不同光照条件下,能准确提取车辆信息,实现精准匹配和跟踪。(3)论文提出一种广泛适用的车辆遮挡检测分割算法。车辆间交互容易造成车辆的遮挡,影响跟踪效果。文中遮挡的处理主要包含两步操作,首先对目标候选框进行遮挡检测,然后对符合条件的遮挡框进行遮挡的分割,得到分割后车辆信息,用于后续车辆跟踪,一定程度上解决了车辆的遮挡问题。根据以上算法,本文选取多种场景下交通道路视频进行检测跟踪仿真实验。通过对车辆的正确跟踪,错误跟踪以及丢失跟踪帧数进行的统计表明,该框架是一个稳健的车辆检测与跟踪系统,可以有效去除阴影以及处理遮挡,从而准确的检测与跟踪车辆。
[Abstract]:With the development of society, Intelligent Transportation system (its) has become the focus of attention in every country in the world. Intelligent Transportation system (its) plays an important role in road monitoring and traffic management. Vehicle detection and tracking is one of the hot and difficult problems in Intelligent Transportation system (its). This paper focuses on the vehicle motion detection, tracking, shadow removal and occlusion processing, and applies these methods to the video surveillance system. The main work is as follows: 1) this paper proposes an improved background modeling algorithm and a shadow removal algorithm based on non-parameterized feature fusion. Firstly, the background modeling algorithm is introduced, and Vibe Visual background extractor is selected as the target detection algorithm of the system. The shortcomings of the Vibe modeling algorithm are described, and the improved method is put forward. This method can effectively suppress the ghost problem which is easily introduced by Vibe algorithm, so as to detect the target more accurately. At the same time, in order to eliminate the influence of moving vehicle shadow on the foreground result, an improved method is proposed to remove the shadow by combining the HSV HueSaturationValue feature with the texture LBPU Local binary pattern feature, which has no inclusion for HSV feature. This fusion algorithm effectively removes the vehicle shadow. In this paper, a particle filter tracking algorithm based on improved observation model is proposed. Firstly, the target tracking algorithm is studied, several typical algorithms are introduced, and the basic principle of the algorithm is described, and a comparative analysis is given. At the same time, a vehicle observation model based on improved hog histogram of oriented gradient is proposed, which can accurately extract vehicle information under different illumination conditions. This paper presents a widely applicable segmentation algorithm for vehicle occlusion detection. The interaction between vehicles can easily cause vehicle occlusion and affect the tracking effect. The processing of occlusion in this paper mainly consists of two steps: firstly, the target candidate is detected, then the suitable occlusion frame is segmented to get the vehicle information after segmentation, which can be used for subsequent vehicle tracking. To some extent, the problem of vehicle occlusion is solved. According to the above algorithms, this paper selects a variety of scene traffic road video detection and tracking simulation experiment. The statistics of correct tracking, error tracking and lost tracking frame count show that the framework is a robust vehicle detection and tracking system, which can effectively remove shadows and deal with occlusion. Thus accurate detection and tracking of vehicles.
【学位授予单位】：北京邮电大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：U495;TP391.41

【相似文献】