CASIA OpenIR  > 毕业生  > 硕士学位论文
Alternative TitleResearch on moving object detection Techniques in smart video surveillance system
Thesis Advisor张文生
Degree Grantor中国科学院研究生院
Place of Conferral中国科学院自动化研究所
Degree Discipline计算机应用技术
Keyword智能视频监控 运动目标检测 多种类视觉特征 混合高斯 定量分析 Smart Video Surveillance Moving Object Detection Multi-category Visual Features Mixture Of Gaussian Model Quantitative Analysis
Abstract如何精确且实时地检测场景中的运动目标是很多智能视频监控系统,诸如安全监控、智能交通、智能火检等系统中的一个非常基础和关键的技术环节。然而,在实际的应用环境中,运动目标检测技术经常面临着很多自然因素和非自然因素造成的技术难题。因此,为了提高智能视频监控系统在整体上的检测性能,提出一种能够解决上述技术难题且稳健鲁棒的运动目标检测算法是非常必要的,也是一项极具挑战的工作。 本文主要针对静止摄像头下的运动目标检测技术进行研究与改进。首先,对当前主流的运动目标检测技术进行归类总结,并分析了各自的优缺点。其中,对以背景减除建模技术为基础的运动目标检测方法进行了重点讨论。这一部分主要是从构建背景场景模型使用的特征种类以及描述特征模型随时间变化的方法两个角度对背景减除建模技术加以详细论述。然后,针对传统背景减除建模技术检测结果中经常出现的前景空洞以及虚警的离散噪音问题,本文提出了一种基于颜色,边缘和纹理的混合视觉特征的背景减除建模技术。新的视觉特征向量是基于8×8分辨率图像块的离散余弦变换域的系数组合得到。为了适应背景场景随时间变化的多模态情形,本文同时采用经典的混合高斯模型对变换域获取的多种类视觉特征进行描述和实时更新。最后,本文的方法与经典的背景减除技术在四组不同场景的视频序列上进行了对比实验。基于准确率和召回率的定量分析结果表明,新的基于多种类视觉特征的背景减除方法有效的解决了传统方法中存在的前景空洞和存在离散虚警噪音点干扰问题,同时也为后继的高层视觉分析任务打下了良好的基础。
Other AbstractHow to detect moving objects in scenes in real time and precisely is a fundamental and critical step in many smart vision systems such as security surveillance, intelligent transportation, smart fire detection, etc. In practical application environments, however, moving object detection methods are often faced with many kinds of problems caused by natural environmental factors and non-natural environmental factors. Therefore, to propose a robust moving object detecting algorithm is a necessary and challenge work. The paper focuses on research and improvement of moving object detection techniques with static cameras condition. Firstly, the moving object detecting method are classified and summarized. Simultaneously, their advantages and disadvantages are also analyzed. We lay stress on discussion about moving object detection methods based on background subtraction modeling techniques. In this section, we mainly discuss background subtraction modeling method in terms of feature categories for constructing background scenes and methods of depicting feature model changes with time. Secondly, we propose a novel background modeling method based on multi-category visual features (i.e. color, texture, edge). The new modeling feature vector is extracted based on coefficients of an 8×8 sized block in its discrete cosine transform domain. To adapt to changes of scenes through time, we also use the classic mixture of Gaussian model to describe multi-category visual feature extracted from the discrete cosine transform domain and update them in real times. Finally, we conduct comparative experiments with traditional method on four different challenging test video sequences. The quantitative analysis data based on precision ratio and recall ratio demonstrates that the new proposed background modeling method utilizing multi-category visual features solves problems of foreground apertures and false positive noises effectively, which will be a baseline for high-level vision analysis tasks.
Other Identifier200728017029250
Document Type学位论文
Recommended Citation
GB/T 7714
常晓夫. 智能视频监控系统中的运动目标检测技术研究[D]. 中国科学院自动化研究所. 中国科学院研究生院,2010.
Files in This Item:
File Name/Size DocType Version Access License
CASIA_20072801702925(1683KB) 暂不开放CC BY-NC-SAApplication Full Text
Related Services
Recommend this item
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[常晓夫]'s Articles
Baidu academic
Similar articles in Baidu academic
[常晓夫]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[常晓夫]'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.