CASIA OpenIR  > 毕业生  > 博士学位论文
基于目标层次结构的车辆检测技术及应用研究
Alternative TitleVehicle Detection Based on Hierarchical Structure and Its Applications
田滨
Subtype工学博士
Thesis Advisor王飞跃
2014-05-25
Degree Grantor中国科学院大学
Place of Conferral中国科学院自动化研究所
Degree Discipline控制理论与控制工程
Keyword智能交通系统 交通视频监控 车辆检测 概率图模型 目标检测语法 复杂系统 Intelligent Transportation System Video-based Traffic Surveillance Vehicle Detection Probabilistic Graph Model Object Detection Grammars Complex System
Abstract车辆检测技术是交通视频监控系统中的一项关键技术。随着城市化进程的不断加速,城市规模不断扩大,汽车的保有量迅速上升,日益繁忙的城市交通系统产生了交通安全、交通拥堵和环境污染等交通问题。为了缓解交通问题,智能交通系统(ITS)近年来得到了世界各国的广泛关注和应用。交通视频监控系统作为ITS的重要组成部分,在缓解交通拥堵、减少交通事故等方面发挥了重要作用。然而,复杂的交通道路环境为车辆检测技术带来了巨大的挑战,比如,拥堵场景下的车辆遮挡、车辆静止,恶劣天气条件,一天中不同时段中的光照条件变化等。如何提高车辆检测技术在复杂交通环境下的检测效果,仍然是一个值得研究的问题。 本论文分析现有的复杂交通系统,针对现有交通监控系统的问题和交通监控的现实需求,从目标层次结构的角度研究车辆检测技术,并实现用于实际交通道路的监控系统。具体来说,本文的主要贡献包括以下几个方面: 1) 分析车辆目标的显著部件特征,研究显著部件的检测问题。在城市交通环境中,车辆检测技术经常面对部分遮挡、类内变化、多姿态、复杂背景等问题。将车辆作为一个整体来检测不再合理,而基于部件(局部特征)的车辆检测方法提供了解决这些问题的可行思路。本论文中选取具有小类内距离和大类间距离的部件(即显著部件,包括车牌和车辆尾灯)来表示车辆。在选取有鉴别力的显著部件之后,通过部件的判别性外观特征对其定位。对于车牌,利用其特有的颜色和梯度特征进行检测;对于车辆尾灯,利用其特有的颜色和区域特征进行检测。 2) 基于概率图模型技术,研究车辆显著部件融合问题。在检测到车辆的显著部件之后,如何根据部件检测结果推导出车辆,是一个关键问题。如果采用刚性结构模型描述部件之间的关系,则对车辆的结构约束太强,难以处理类内变化、多姿态等实际问题。本论文研究车辆的松散结构模型,灵活地描述部件之间的关系,采用一种无向概率图模型――马尔科夫随机场(MRF)来融合利用部件之间的上下文信息,实现可靠的部件融合,提高车辆检测的准确性。 3) 分析交通场景和车辆目标的特性,研究车辆部件的层次划分问题。在拥堵的交通场景中,车辆与其他交通目标之间经常存在遮挡,遮挡问题一直是车辆检测的难点之一。其主要原因是被遮挡的车辆丢失了部分外观信息,导致车辆检测器难以根据可见信息来判断车辆的位置。此外,常用的目标检测模板通常为矩形模板,而车辆具有不同的姿态和不同的车型,因而其外观形状未必是矩形,因此利用矩形模板检测车辆目标,很容易包含大量的背景和其他目标的信息。基于以上观察,我们提出了一种双层结构的部件划分方法,第一层为语义部件,根据车辆目标的特性进行选取;第二层为子部件,根据语义部件的部件模型自动获取。这种划分方式为不同车型和不同姿态下的车辆采用了部件共享的策略,因此可以在一定程度上减少部件数量,从而降低模型复杂度。这种划分方式中的部件尽量覆盖车辆本身,而排除背景和其他目标的影响,既可以有效表示车辆目标,又有利于处理遮挡问题。 4) 基于分层结构的车辆部件划分方式,研究车辆检测语法的训练和...
Other AbstractVehicle detection is a key technology in video-based traffic surveillance system. With the accelerating process of the urbanization and city scale expanding unceasingly, the recoverable amount of the automobile is also rising rapidly. Therefore, city traffic system becomes increasingly busy, resulting in traffic problems such as traffic safety, traffic congestion and environment pollution etc. In order to alleviate the traffic problems, in recent years, intelligent transportation systems (ITS) attract lots of attention throughout the world. Video-based traffic surveillance system, as an important part of ITS, plays an important role in alleviating traffic congestion, reducing the traffic accidents. However, the complex traffic environments have brought about huge challenges to vehicle detection, including vehicle occlusion and stationary vehicle under congestion scenarios, lighting variation under bad weathers and at different times of day. Aiming at solving the problems of existing traffic surveillance system and meeting the new demands for traffic surveillance, this thesis analyses the existing complex traffic system, studies vehicle detection technology from the point of view of object hierarchical structure and realizes the traffic surveillance system for the real traffic. Specifically, the main contributions of this thesis include the following aspects: 1) We analyze the salient parts of vehicle object and study the detection of these salient parts based on their distinctive features. In the urban traffic environment, vehicle detection technology often faces problems such as partial occlusion, intra-class variation, pose variation, background clutter and so on. It is not reasonable to detect the vehicle as a whole, while the part (local features) based vehicle detection method provides feasible ways to solve these problems. This thesis selects vehicle parts with small intra-class difference and large inter-class difference (i.e., significant parts) to represent the vehicle, including vehicle license plate and rear-lamps. After selecting the salient parts, we need to locate them through the discriminative appearance features. The vehicle license plate is detected based on its unique color and gradient features; as for vehicle rear-lamps, they are detected based on their unique color and regional characteristics. 2) Based on the probabilistic graphical model technology, we study the fusion of salient parts for vehicle detection. After detecting t...
shelfnumXWLW2001
Other Identifier201118014628019
Language中文
Document Type学位论文
Identifierhttp://ir.ia.ac.cn/handle/173211/6601
Collection毕业生_博士学位论文
Recommended Citation
GB/T 7714
田滨. 基于目标层次结构的车辆检测技术及应用研究[D]. 中国科学院自动化研究所. 中国科学院大学,2014.
Files in This Item:
File Name/Size DocType Version Access License
CASIA_20111801462801(5152KB) 暂不开放CC BY-NC-SAApplication Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[田滨]'s Articles
Baidu academic
Similar articles in Baidu academic
[田滨]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[田滨]'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.