监控系统预警视频的分布式检索设计与实现

Design and Implementation of Distributed Retrieval in Video Monitoring System

摘要:

视频监控系统在日积月累的监控过程中会产生大量的视频信息,如何在海量视频中基于内容快速检索出需要的信息成为亟待解决的问题;文章在视频检索过程中,将视频图像的SIFT特征提取匹配与MapReduce并行模式相结合,同时利用LSH映射对视频图像进行分组并存储在HBase数据库中,在配合并行分析的同时减少了计算量;最后实验证明,采用MapReduce并行模式提取图像SIFT特征值,时间会随着集群节点的增加而不断降低,并最终趋于稳定;由此可知,对基于内容的视频检索采用并行化分析可以显著提高效率,实现了Hadoop架构在视频分析方面的应用拓展。

Abstract:

Video surveillance system will produce a lot of video information in the monitoring process over time,how to search needed information quickly,which based on content,in the massive videos has become an urgent problem to be solved. This paper,during the video retrieval process,combines the SIFT feature extraction and matching of the video image with the MapReduce parallel mode,and meanwhile uses the LSH mapping to divide the video image into groups and stores it in the HBase database ,which reduces the amount of calculation in conjunction with the parallel analysis. Finally,the experiments show that by using MapReduce parallel mode to extract the SIFT extraction characteristic value of image,time will continue to decrease with the increase of the cluster nodes,and finally tend to be stable. Therefore,to the video retrieval based on content,it can improve the efficiency remarkably by using the parallel analysis,and it realizes the expansion of the application of Hadoop architecture in the video analysis.