相机运动条件下的视频车辆检测
发布时间：2018-08-05 20:30
【摘要】：视频车辆检测是一种在视频序列中提取运动车辆对象的技术,其广泛运用于视频监控、智能交通检测等系统中。由于运动车辆检测技术,特别是相机运动条件下的视频车辆检测,具有复杂性、多变性等特点,该技术仍处于起步阶段,需要不断的研究并加以改进。为了在相机运动条件下准确检测出车辆,本文围绕着以下几个方面来开展研究工作:(1)对本文涉及到的理论基础知识进行学习研究。总结概括涉及视频图像处理,尤其是视频目标检测等相关国内外文献资料;筛选并学习基于相机运动条件下视频车辆检测的相关知识,重点是近几年关于动摄像头下视频车辆检测研究成果;针对动摄像头视频车辆检测研究现状,并结合自身知识积累,构建本文算法框架。(2)对视频车辆检测方法进行分析研究。视频车辆检测方法可根据摄像头是否运动进行分类,本论文简单介绍了静摄像头下视频运动车辆检测方法,具体介绍相机运动条件下的视频车辆检测方法。(3)对相机运动条件下的视频车辆检测进行算法设计。算法分为四部分。第一部分全局运动估计与补偿算法,并对基础算法进行说明,在这基础上,通过分析不同的全局运动估计方法的优缺点,提出使用六参数仿射变换模型,估计仿射变换参数,然后对仿射变换后的图像进行补偿。第二部分高斯差分算法,改进了运动补偿后的差分步骤,以便获得更好的检测效果。第三部分非参数核密度估计算法,以此对检测进行优化。第四部分使用矩形框对检测出的车辆进行目标定位。(4)对本论文的算法进行实验验证。利用VS2010和Matlab软件平台,并结合OpenCV开源库,编写本算法的实验仿真程序,以动摄像头拍摄的视频为输入,进行实验。实验验证过程分为两部分:第一部分实验用来验证本文算法的功能实现;第二部分实验是对比试验,用来验证本文算法的准确性和高鲁棒性。研究创新有两点:(1)为了减小摄像头运动对视频中运动目标检测的影响,提高运动估计的准确度,在进行仿射运动估计时,对目标帧的前后帧采用不同的仿射变换矩阵,计算量降低了,仿射变换效果提高了。(2)使用非参数核密度估计对得到的检测目标进行优化,减少各环节带来的目标空洞问题和噪声灵敏度问题的影响。研究的不足是,仿射运动估计参数的获取,计算量较大,时间较长,仿射运动估计参数的获取速度仍有待提高,算法的天气情况适应性,需要进一步研究及改进。
[Abstract]:Video vehicle detection is a technique for extracting moving vehicle objects in video sequence. It is widely used in video surveillance and intelligent traffic detection systems. Because of the characteristics of vehicle detection technology, especially video vehicle detection under camera motion conditions, it is still in its infancy. In order to detect the vehicle accurately under the camera movement, this paper focuses on the following aspects: (1) study and study the theoretical basic knowledge involved in this article. Summarize and summarize the related domestic and foreign documents related to video image processing, especially the visual frequency target detection. And learn the related knowledge of video vehicle detection based on camera motion. The focus is on the research results of video vehicle detection in recent years. In view of the current situation of video vehicle detection research in mobile camera, and combining with the knowledge accumulation, this paper constructs the algorithm framework. (2) video vehicle detection methods are analyzed and studied. The vehicle detection method can be classified according to the motion of the camera. This paper briefly introduces the video moving vehicle detection method under the static camera and introduces the video vehicle detection method under the camera motion condition. (3) the algorithm is designed for the video vehicle detection under the camera motion condition. The algorithm is divided into four parts. The first part is global. The motion estimation and compensation algorithm and the basic algorithm are explained. On this basis, by analyzing the advantages and disadvantages of different global motion estimation methods, the six parameter affine transform model is used to estimate the affine transformation parameters, and then the affine transform image is compensated. The second part of the Gauss difference algorithm is improved after the motion compensation. The third part of the non parametric kernel density estimation algorithm is used to optimize the detection. The fourth part uses the rectangle frame to target the detection of the vehicle. (4) the experimental verification of the algorithm in this paper. Using the VS2010 and Matlab software platform, and combining the OpenCV open source library, write this The experimental simulation program of the algorithm is carried out with the video taken by the moving camera as input, and the experiment is divided into two parts: the first part is used to verify the function realization of the algorithm. The second part of the experiment is a contrast test, which is used to verify the accuracy and robustness of the algorithm. (1) to reduce the perturbation. Like the effect of head motion on moving target detection in video, the accuracy of motion estimation is improved. When carrying out the affine motion estimation, different affine transform matrices are used for the front and back frames of the target frame. The amount of computation is reduced and the effect of affine transformation is improved. (2) the detection targets are optimized by using non parametric kernel density estimation, and each of them is reduced. The problem of the target cavitation and the noise sensitivity caused by the link. The shortage of the research is that the parameters of the affine motion estimation are obtained, the computation is large and the time is long. The acquisition speed of the affine motion estimation parameters remains to be improved, and the weather conditions of the algorithm are adaptable, and further research and improvement are needed.
【学位授予单位】：山东师范大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41
本文编号：2166930
[Abstract]:Video vehicle detection is a technique for extracting moving vehicle objects in video sequence. It is widely used in video surveillance and intelligent traffic detection systems. Because of the characteristics of vehicle detection technology, especially video vehicle detection under camera motion conditions, it is still in its infancy. In order to detect the vehicle accurately under the camera movement, this paper focuses on the following aspects: (1) study and study the theoretical basic knowledge involved in this article. Summarize and summarize the related domestic and foreign documents related to video image processing, especially the visual frequency target detection. And learn the related knowledge of video vehicle detection based on camera motion. The focus is on the research results of video vehicle detection in recent years. In view of the current situation of video vehicle detection research in mobile camera, and combining with the knowledge accumulation, this paper constructs the algorithm framework. (2) video vehicle detection methods are analyzed and studied. The vehicle detection method can be classified according to the motion of the camera. This paper briefly introduces the video moving vehicle detection method under the static camera and introduces the video vehicle detection method under the camera motion condition. (3) the algorithm is designed for the video vehicle detection under the camera motion condition. The algorithm is divided into four parts. The first part is global. The motion estimation and compensation algorithm and the basic algorithm are explained. On this basis, by analyzing the advantages and disadvantages of different global motion estimation methods, the six parameter affine transform model is used to estimate the affine transformation parameters, and then the affine transform image is compensated. The second part of the Gauss difference algorithm is improved after the motion compensation. The third part of the non parametric kernel density estimation algorithm is used to optimize the detection. The fourth part uses the rectangle frame to target the detection of the vehicle. (4) the experimental verification of the algorithm in this paper. Using the VS2010 and Matlab software platform, and combining the OpenCV open source library, write this The experimental simulation program of the algorithm is carried out with the video taken by the moving camera as input, and the experiment is divided into two parts: the first part is used to verify the function realization of the algorithm. The second part of the experiment is a contrast test, which is used to verify the accuracy and robustness of the algorithm. (1) to reduce the perturbation. Like the effect of head motion on moving target detection in video, the accuracy of motion estimation is improved. When carrying out the affine motion estimation, different affine transform matrices are used for the front and back frames of the target frame. The amount of computation is reduced and the effect of affine transformation is improved. (2) the detection targets are optimized by using non parametric kernel density estimation, and each of them is reduced. The problem of the target cavitation and the noise sensitivity caused by the link. The shortage of the research is that the parameters of the affine motion estimation are obtained, the computation is large and the time is long. The acquisition speed of the affine motion estimation parameters remains to be improved, and the weather conditions of the algorithm are adaptable, and further research and improvement are needed.
【学位授予单位】：山东师范大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41
【参考文献】
相关期刊论文 前10条
1 高勇钢;;改进帧差法和背景差法的多目标跟踪[J];巢湖学院学报;2013年06期
2 王斌;何中市;伍星;贾媛媛;;基于高斯金字塔的图像运动估计算法[J];计算机工程与应用;2015年07期
3 栾庆磊;赵为松;;动背景下帧差分法与边缘信息融合的目标检测算法[J];光电工程;2011年10期
4 刘彬;严京旗;施鹏飞;;高斯差分的AdaBoost车牌定位方法[J];智能系统学报;2010年06期
5 孙剑芬;;基于高斯核密度估计的运动目标检测新方法[J];计算机技术与发展;2010年08期
6 付青青;张春海;;高斯模糊图像的复原处理与研究[J];长江大学学报(自然科学版)理工卷;2010年02期
7 盛旭锋;朱方文;李校祖;庄俊;;基于三帧时间差分法的独居老人运动检测[J];计算机工程与应用;2010年13期
8 王久阳;何广军;朱福军;;基于线性高斯滤波的多传感器管理算法[J];探测与控制学报;2009年05期
9 李宁;黄山;张先震;李秀君;;基于背景差分的人体运动检测[J];微计算机信息;2009年21期
10 郑雅羽;田翔;陈耀武;;基于运动矢量对消和差分原理的快速全局运动估计[J];电子与信息学报;2009年04期
相关硕士学位论文 前2条
1 张金花;运动目标跟踪算法的研究[D];兰州大学;2009年
2 方颖;运动目标的检测、定位与跟踪研究[D];山东大学;2008年
，本文编号：2166930
本文链接：https://www.wllwen.com/kejilunwen/ruanjiangongchenglunwen/2166930.html
