CASIA OpenIR  > 毕业生  > 硕士学位论文
视角不变的人的行为分析与理解
Alternative TitleView invariant human action analysis and understanding
张叶银
Subtype工学硕士
Thesis Advisor黄凯奇
2010-06-07
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模式识别与智能系统
Keyword视角不变 运动特征 人的行为识别 时间序列分析 Motion Feature View Invariant Human Action Time Series Analysis
Abstract视频中人的行为分析是计算机视觉中经典且富于挑战性的问题。计算机视觉研究图像信息到高层语义的映射问题,人的行为分析则是挖掘图像序列数据中人的行为的潜在模式,让计算机理解特定场景下人的行为。视频中人的行为分析与理解涉及到数据采集及预处理、特征提取和表达、机器学习和数据挖掘以及系统应用等问题。在计算机视觉发展的短暂历程中,这些问题的解决方法也层出不穷,促进了计算机视觉学科的发展,并在技术的推动作用下,形成了初具规模的产业链与新的市场需求。但是,现有方法还不够成熟,实际应用中也遇到了种种限制,该领域需要更深入地研究。 本文围绕人的行为分析,在智能视觉监控系统的应用背景下,以解决实际问题为向导,从理论上和实际应用上进行探索。本文内容主要涉及视角不变性、运动特征提取和表达、序列数据的分析与建模等问题,特别是针对视角变化问题进行了深入的研究。 本文主要工作可以分为以下几个部分: 提出一种基于几何不变量的特征提取算子,该算子是以人的运动轨迹为前提,直接从图像中提取视角不变量,在不需要任何三维信息的条件下,识别不同视角下人的行为,有效克服视角变化带来图像投影失真从而导致识别性能下降的问题。 提出一种基于运动特征的人的行为识别方法,并融合视角不变的特征,提高对视角变化的鲁棒性。该方法利用时空关键点检测人的运动,提取图像中局部区域运动特征,并由时空关键点构造出视角不变特征。采用基于最大熵原理的判别式模型---条件随机场融合运动特征和视角不变的特征,并进行时间序列分析和建模。实验验证了该方法在视角变化情况下具备一定的鲁棒性。 针对智能视觉监控系统功能上的需求,结合本文的研究,将运动特征提取和识别的方法应用到面向家庭安全的智能监控系统中,提高系统在特定场景下对人的某些特定行为的识别性能 总之,本文针对理论和实际应用中存在的问题,在人的行为分析与理解的研究上作出了有益的探索。
Other AbstractHuman action recognition in video sequences is a most important and challenging problem in computer vision, which aims to build the mapping between image information and semantic understanding of sequential data. While the analysis and recognition of human action tries to discover the underlying patterns of human action in image data, it could be also useful in many real applications such as intelligent video surveillance, content based image retrieval, event detection and so on. Beyond image processing, human action recognition involves a series of problems, like image data acquisition, feature extraction and representation, machine learning and data mining and application problems that may come out in system running. In the short history of development of computer vision, scientists have worked out many good solutions of these problems. And computer vision products have become indispensable in market as the demand increases. However, existing technologies haven't come to the stage that we have ever dreamed of, because we have only moved out one little step in this area and encountered difficulties and limitations. Research in this area is still worth our close attention. In this thesis, we made some new attempts on human action recognition about some real application problems of video surveillance system. The work involves in motion features extraction and representation and sequential data analyzing and modeling, especially in action recognition with view change. The main contributions of this thesis include following issues: we propose a new descriptor for view invariant feature extraction based on geometry invariance in computer vision. The descriptor is based on human action trajectories so that we could compute invariants directly from image data under different cameras. It does not need any 3D information of the image and could help to solve ineffectiveness caused by the projective distortion of view change. we proposed a new method of human action recognition based on motion feature. In the step of detection, we apply space-time interest point to detect the motion in image. In recognition, we model the sequential motion feature using a more expressive discriminative model. We tested the method on multi-view action database and achieved better results. we design a video surveillance system characterized by motion feature extraction and recognition techniques. The system could be used for intrusio...
shelfnumXWLW1538
Other Identifier200728014628048
Language中文
Document Type学位论文
Identifierhttp://ir.ia.ac.cn/handle/173211/7545
Collection毕业生_硕士学位论文
Recommended Citation
GB/T 7714
张叶银. 视角不变的人的行为分析与理解[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_20072801462804(6014KB) 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
[张叶银]'s Articles
Baidu academic
Similar articles in Baidu academic
[张叶银]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[张叶银]'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.