摘要
目前,基于Hadoop视频处理的传统方法都是通过MapReduce从本地文件系统读取数据,利用帧字节流进行MapReduce间数据传输,这会产生大量的系统IO,造成系统资源浪费;针对此问题,提出一种基于Hadoop平台的视频处理方法,实现Hadoop支持的视频类型扩展,设计了MapReduce相关视频数据处理接口,使Hadoop可以更快速处理视频文件;通过在多台计算机组成的集群实验表明,该方法在运行时间上比传统方法缩短10%,IO读写量减少50%以上,提升了Hadoop视频文件的处理效率。
At present, traditional methods based on Hadoop video processing read data from local file system through the MapReduce. Using byte--stream, they transfer data between MapReduce, but this will produce a large amount of system IO and cause loss of resources. Considering the issue, this passage proposes a video processing method based on Hadoop platform. The method realizes video type extensions supported by Hadoop and designs video data processing interfaces about MapReduce, which make Hadoop process video files much faster. Experiments on computer cluster illustrate that compared with traditional methods, this method~ s running time decreases by 10 % and the amount of IO read and write decline more than 50%. So it improves the processing efficiency of Hadoop video files
出处
《计算机测量与控制》
2015年第12期4117-4120,共4页
Computer Measurement &Control
基金
国家自然科学基金项目(61172181)
