CASIA OpenIR  > 毕业生  > 硕士学位论文
Alternative TitleOnline Content-aware Video Condensation
Thesis Advisor李子青
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视频摘要 粘连式跟踪 视频索引 视频监控 Video Synopsis Sticky Tracking Video Indexing Video Surveillance
Abstract随着世界范围内对社会公共安全和公众保护需求的日益增长,数以万计的监控摄像头被安装在公园、体育场馆、大型广场、学校、医院、商业街、住宅小区等公众活动和聚集的场所,有效浏览、检索以及存储这些海量监控视频是一个极大的挑战。 视频浓缩技术应运而生。该技术通过对运动物体序列在时间轴上进行重排,以达到缩短原始视频长度的目的。但是,传统视频浓缩算法采用离线处理模式,具有硬件需求和算法复杂度双高等缺点,不可实时使用。针对传统浓缩技术固有的缺点,本文提出在线内容感知视频浓缩算法。受俄罗斯方块游戏的启发,该算法采用在线处理模式对输入视频进行浓缩。本文主要的工作和贡献有: 1) 提出了一种在线内容感知浓缩框架。该框架能够不间断处理24×7的监控视频,并且实时不间断地生成浓缩浓缩。与传统方法中浓缩率是由用户指定不同,该内容感知框架使得本文算法具有自适应的浓缩率。 2) 提出了一种粘连式跟踪算法,该算法有效地消除了传统跟踪算法应用于视频浓缩中产生的频闪现象,大大提升了浓缩视频的视觉效果。 3) 提出了一种运动物体序列在线填充方法,主要包括浓缩空间的构建、阶梯式能量优化以及重分配策略。 4) 提出了一种在线主背景选择算法,目的在于生成浓缩视频中的背景序列。与以往方法相比,该算法具有速度快、消除消耗低等优点。 5) 提出了在线视频浓缩的硬件加速方法,使得最终算法能以超实时的速度运行。 6) 实现了一个在线内容感知视频浓缩原型系统。除了在线浓缩的基本功能外,该系统还有一个界面友好的回放子系统。 总而言之,本文的在线内容感知视频浓缩算法具有更快的速度、消耗更低的内存、拥有自适应的浓缩率以及更好的视觉效果。作者相信在线内容感知视频浓缩算法有很大潜力影响下一代的监控视频的存储与浏览。
Other AbstractAs the demand grows worldwide for the public security and protection, tens of thousands of surveillance cameras are installed in the places of public activities, such as park, stadium, square, school, hospital, business street, residential block and so on. This presents formidable challenges to the browsing, retrieval and storage of the massive surveillance videos. The technology of video condensation (also called video synopsis) emerges at the proper moment. Such technology shortens the video by rearranging moving object sequences along the time axis. However, the current techniques are offline methods, and are expensive in time and space, by which they are prevented from actual use. Therefore, we propose an online content-aware video condensation solution to overcome the problems of traditional methods. Inspired by the game Tetris, our method adopts an online condensation approach. The main works and contributions of this thesis are: 1) Propose an online content-aware video condensation framework. Such framework has a natural-born ability to take 24×7 endless surveillance video and generate endless condensed video. The content-aware framework makes our method has a self-adapting condensation ratio, and this differs from traditional methods in which the condensation ratio is specified by the user. 2) Propose a novel tracking method, called sticky tracking, to eliminate the undesirable blinking effect arised from traditional tracking methods. The visual effect of condensed videos increases dramatically by the new tracking algorithm. 3) Propose an online tube filling method, including the construction of condensed space, stepwise energy optimization and reallocation strategy. 4) Propose an online principal background selection method for the purpose of generating the background sequence of condensed video. Compared to existing methods, the proposed method has advantages of high speed and low memory consumption. 5) Propose an acceleration strategy, which makes the whole algorithm achieve faster-than-real-time video condensation. 6) Implement an online content-aware video condensation prototype system. Apart from the basic function of video condensation, it has a friendly ``playback'' subsystem. In brief, the proposed online content-aware video condensation algorithm has a much higher speed, much lower memory consumption, self-adapting condensation ratio and better visual effect. The author believes that the proposed solution and algorithms...
Other Identifier200928014628033
Document Type学位论文
Recommended Citation
GB/T 7714
冯仕堃. 在线内容感知视频浓缩研究[D]. 中国科学院自动化研究所. 中国科学院研究生院,2012.
Files in This Item:
File Name/Size DocType Version Access License
CASIA_20092801462803(10180KB) 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
[冯仕堃]'s Articles
Baidu academic
Similar articles in Baidu academic
[冯仕堃]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[冯仕堃]'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.