SIA OpenIR  > 自动化系统研究室
基于视觉的交通监控技术研究
Alternative TitleResearch on the Vision based Traffic Monitoring Techniques
隋晔1,2
Department自动化系统研究室
Thesis Advisor马钺
ClassificationU491
Keyword视觉 交通监控
Call NumberU491/S95/2003
Pages82页
Degree Discipline模式识别与智能系统
Degree Name硕士
2003-07-02
Degree Grantor中国科学院沈阳自动化研究所
Place of Conferral沈阳
Abstract基于视觉的交通监控系统是计算机视觉技术在智能交通系统中的主要应用之一。计算机视觉为交通监控系统提供了更为直观方便的手段,交通环境中大量信息来自于视觉,用计算机视觉技术来处理或理解这类信息是一种自然的选择。它对于缓解交通阻塞,提高道路通过率,减少事故的发生以及加强交通安全具有重要的现实意义。 本文从交通监控系统的低层视觉技术到中高层视觉展开了探讨,比较系统地研究了交通监控系统中的几个关键问题。运动视觉分析技术和数字图像处理技术是基于视觉的交通监控系统中的关键所在。因此我们首先介绍了动态图像分析的重要步骤—运动分割,详细分析了运动分割的主要方法和存在的问题,然后介绍了动态过程的分析技术—运动跟踪。接着介绍了数字图像处理技术的一些基本理论,针对交通场景中对比度较低的图像,对有关增强图像对比度和边缘的算法进行了必要的改进和实验仿真。提出了一种基于概率统计的运动目标检测算法,该算法通过建立统计模型,来自动判断是由噪声引起的象素点变化还是目标运动引起的变化,其定位能力强大,对于由摄像机振动和天气﹑光线等引起的噪声有很好的鲁棒性。在运动检测的基础上,详细介绍了基于连续时间限制和最大可能性估计的目标分类机制和分类过程。 交通参数检测是监控系统中的一个重要环节,本文提出了一种基于图像帧差和虚拟线状态检测的检测方法,该方法通过对交通图像中车辆和场景信息进行图像分割和符号分析来有效地检测交通参数。实验结果显示该方法简单有效,能实时获取各种交通信息,对车辆计数和车辆速度检测的准确率都很高,并具有满意的实时特性。
Other AbstractVision based Traffic Monitoring System is one of the main applications in Intelligent Transportation Systems. Vision is the main source of information in traffic scenes, so it is a natural way of applying information technologies such as image processing, pattern recognization and intelligent control to process and understand this kind of information. It has realistic significance for relaxing traffic congestion, improving passing efficiency of road, decreasing traffic accidents, and strengthening traffic safety. This thesis is focused on the vision techniques from low-level to middle and high level in traffic monitoring system, brings forward effective methods regarding several key problems. A lot of techniques for motion vision analysis and digital image processing are applied in vision-based traffic monitoring surveillance. At present, accurate motion estimation is still a tough problem to solve, so methods on motion segmentation are discussed in detail at first. Then introduced the knowledge of motion tracking and the basic theories of digital image processing. Due to the fact that the contrast of video frames in the Surveillance System is relatively low, we have necessary improvement and experiment simulation on the operators about expanding the contrast and edges of surveillance images. In the process of moving target detection, it is difficult to determine the thresholds that can get rid of noise and preserve the motion information at the same time. A method based on probability statistics was presented to dynamically change the threshold by taking into account the statistical behavior of each pixel's neighborhood. The experiments results showed that more accurate moving detection could be obtained by using this algorithm. Also we introduced the process of target classifying, the principle of temporal consistency and maximum likelihood estimation were employed to realize classification. Traffic parameters' detection is an important step in traffic monitoring system. This paper presents a method based on moving target detection and virtual line detection; it can effectively acquire traffic parameters by analyzing the vehicle and scene information extracted from surveillance images. Using this method we could obtain the vehicle count, vehicle speed, queue length and queue state, and experiment results showed that the correct rates of vehicle count and vehicle speed detection are high and the real-time performance is satisfactory.
Language中文
Contribution Rank1
Document Type学位论文
Identifierhttp://ir.sia.cn/handle/173321/9443
Collection自动化系统研究室
Affiliation1.中国科学院沈阳自动化研究所
2.中国科学院研究生院
Recommended Citation
GB/T 7714
隋晔. 基于视觉的交通监控技术研究[D]. 沈阳. 中国科学院沈阳自动化研究所,2003.
Files in This Item:
File Name/Size DocType Version Access License
基于视觉的交通监控技术研究.pdf(665KB) 开放获取LicenseApplication Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[隋晔]'s Articles
Baidu academic
Similar articles in Baidu academic
[隋晔]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[隋晔]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.