基于Haar-like矩形特征的视频车辆检测算法研究与实现
发布时间：2019-06-25 21:31
【摘要】：智能交通管理系统是目前21世纪道路交通管理的发展趋势。高速公路的不断快速发展和车辆管理体制的不断完善,为智能交通管理系统进入实际应用领域提供了契机。基于视频的智能交通管理系统能检测出很多交通参数,其中一个重要参数就是车流量。因此基于视频的交通车流量检测技术成为了智能交通管理系统的重要研究领域。 本文根据实际的交通状况,以车辆交通图片为研究对象,重点研究的是实现一种实时道路车辆识别方法,即如何快速、准确地在复杂背景的图像中识别出所有目标车辆的检测。在总结和分析现有的车辆检测的基础上,针对视频图像的预处理、车辆的检测等问题进行了探索和研究,提出一种帧差法和Haar-like特征相结合的车辆检测的算法,并通过实验证明了新方法的有效性。 本文在现有的研究基础上,分析了Haar-like矩形特征的形式及利用积分图像法快速计算方法。并采用AdaBoost算法训练分类器及采用了12种矩形特征来表示车辆特征,并对车底阴影部分用单特征表示,并将车辆的两个垂直边缘与车底阴影部分组成“U”形特征。 本文根据车辆特征识别的不同要求,在VC平台下调用OpenCV标准库函数应用C语言对识别的要求进行了编程实现。并把该算法应用到实时道路上采集的交通车辆图像中进行了相应的实验测试：车流量检测。并且根据测试实验的内容和结果分别给出在车辆外形多样性和外界环境复杂性下的实验统计数据和性能评价。
[Abstract]:Intelligent traffic management system is the development trend of road traffic management in the 21 st century. The rapid development of expressway and the continuous improvement of vehicle management system provide an opportunity for intelligent traffic management system to enter the field of practical application. Video-based intelligent traffic management system can detect many traffic parameters, one of which is traffic flow. Therefore, video-based traffic flow detection technology has become an important research field of intelligent traffic management system. According to the actual traffic situation, this paper takes the vehicle traffic picture as the research object, and focuses on how to realize a real-time road vehicle recognition method, that is, how to identify all the target vehicles quickly and accurately in the complex background image. On the basis of summing up and analyzing the existing vehicle detection, this paper explores and studies the problems of video image preprocessing and vehicle detection, and proposes a vehicle detection algorithm which combines frame difference method with Haar-like features, and the effectiveness of the new method is proved by experiments. In this paper, on the basis of the existing research, the form of Haar-like rectangular features and the fast calculation method by integral image method are analyzed. The AdaBoost algorithm is used to train the classifier and 12 kinds of rectangular features are used to represent the vehicle features, and the shadow part of the bottom of the vehicle is represented by a single feature, and the two vertical edges of the vehicle and the shadow part of the bottom of the vehicle are formed into "U" features. In this paper, according to the different requirements of vehicle feature recognition, the recognition requirements are programmed with OpenCV standard library function and C language on VC platform. The algorithm is applied to the traffic vehicle image collected on the real-time road to carry out the corresponding experimental test: traffic flow detection. According to the content and results of the test experiment, the experimental statistical data and performance evaluation under the condition of vehicle shape diversity and external environment complexity are given respectively.
【学位授予单位】：昆明理工大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP391.41;U495
本文编号：2506027
[Abstract]:Intelligent traffic management system is the development trend of road traffic management in the 21 st century. The rapid development of expressway and the continuous improvement of vehicle management system provide an opportunity for intelligent traffic management system to enter the field of practical application. Video-based intelligent traffic management system can detect many traffic parameters, one of which is traffic flow. Therefore, video-based traffic flow detection technology has become an important research field of intelligent traffic management system. According to the actual traffic situation, this paper takes the vehicle traffic picture as the research object, and focuses on how to realize a real-time road vehicle recognition method, that is, how to identify all the target vehicles quickly and accurately in the complex background image. On the basis of summing up and analyzing the existing vehicle detection, this paper explores and studies the problems of video image preprocessing and vehicle detection, and proposes a vehicle detection algorithm which combines frame difference method with Haar-like features, and the effectiveness of the new method is proved by experiments. In this paper, on the basis of the existing research, the form of Haar-like rectangular features and the fast calculation method by integral image method are analyzed. The AdaBoost algorithm is used to train the classifier and 12 kinds of rectangular features are used to represent the vehicle features, and the shadow part of the bottom of the vehicle is represented by a single feature, and the two vertical edges of the vehicle and the shadow part of the bottom of the vehicle are formed into "U" features. In this paper, according to the different requirements of vehicle feature recognition, the recognition requirements are programmed with OpenCV standard library function and C language on VC platform. The algorithm is applied to the traffic vehicle image collected on the real-time road to carry out the corresponding experimental test: traffic flow detection. According to the content and results of the test experiment, the experimental statistical data and performance evaluation under the condition of vehicle shape diversity and external environment complexity are given respectively.
【学位授予单位】：昆明理工大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP391.41;U495
【参考文献】
相关期刊论文 前5条
1 文学志;方巍;郑钰辉;;一种基于类Haar特征和改进AdaBoost分类器的车辆识别算法[J];电子学报;2011年05期
2 李良华;罗彬杰;;针对特定目标提取的二值化算法[J];计算机科学;2009年12期
3 王哲;常发亮;;一种基于立体视觉的运动目标检测算法[J];计算机应用;2006年11期
4 王圣男,郁梅,蒋刚毅;智能交通系统中基于视频图像处理的车辆检测与跟踪方法综述[J];计算机应用研究;2005年09期
5 邵平;杨路明;;基于模板分解和积分图像的快速Kirsch边缘检测[J];自动化学报;2007年08期
相关博士学位论文 前2条
1 许言午;面向行人检测的组合分类计算模型与应用研究[D];中国科学技术大学;2009年
2 唐东明;聚类分析及其应用研究[D];电子科技大学;2010年
，本文编号：2506027
本文链接：https://www.wllwen.com/kejilunwen/jiaotonggongchenglunwen/2506027.html
