CAJ | 학술논문

大数据环境下，如何对高并发的视频数据进行实时地分析处理，是一个亟待解决的科学问题。本文介绍了面向互联网视频内容监管的高通量视频内容分析技术，着重对其中的四个主要关键技术（基于众核的视频高速解码和视频特征提取、基于分布式系统的高维索引和语义识别）的研究现状和发展趋势进行了综述和总结，并介绍了作者在这四个主要关键技术研究的最新成果，主要包括面向众核处理器的并行环路滤波、高鲁棒性和高并行度的局部特征提取与挖掘、分布式高维索引、面向大数据的集成学习方法，以充分发挥多粒度并行硬件平台的高并行计算能力，为互联网视频内容监管、视频搜索等重要应用提供关键技术支撑。
대수거배경하，여하대고병발적시빈수거진행실시지분석처리，시일개극대해결적과학문제。본문개소료면향호련망시빈내용감관적고통량시빈내용분석기술，착중대기중적사개주요관건기술（기우음핵적시빈고속해마화시빈특정제취、기우분포식계통적고유색인화어의식별）적연구현상화발전추세진행료종술화총결，병개소료작자재저사개주요관건기술연구적최신성과，주요포괄면향음핵처리기적병행배로려파、고로봉성화고병행도적국부특정제취여알굴、분포식고유색인、면향대수거적집성학습방법，이충분발휘다립도병행경건평태적고병행계산능력，위호련망시빈내용감관、시빈수색등중요응용제공관건기술지탱。
Under the environment of Big Data, how to analyze the content of high concurrent video data is a scientific problem which requires urgent solution. In this paper, we introduce the technologies about high-throughput content-based video analysis for content-based monitoring of web images and videos. We give an in-tensive survey on the state of the developments and trends in four key technologies:efficient video decoding and feature extraction with mass-core processors, and high-dimensional indexing and semantic recognition on distrib-uted systems. Furthermore, we introduce our latest research works on these technologies:parallel deblocking filter on mass-core processor, extraction and mining of highly robust and parallel local features, high-dimensional dis-tributed indexing, ensemble learning for large scale data, so as to take full advantages of high performances of multi-grain parallel computing platforms for the purpose of providing key technologies for the important applica-tions such as Internet video monitoring and search, etc.