高通量视频内容分析技术

High-Throughput Content-Based Video Analysis Technologies

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。

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.