基于DSMI算法和GPU加速的大规模视频检索系统  

Large-scale video retrieval system based on DSMI algorithm and GPU acceleration

作　　者：钟煜[1] 朱敏[1] 陈雪[1] 高宇[1] 

机构地区：[1]四川大学计算机学院机器智能实验室,成都610065

出　　处：《计算机应用研究》2011年第11期4373-4375,共3页Application Research of Computers

基　　金：国家"863"计划资助项目(2008AA01Z119)

摘　　要：本视频检索系统围绕如何处理大规模数据和在线实时检索来展开研究。基于互信息量的镜头边界检测算法检索性能较好,但计算量较大,针对此问题,提出了一种基于动态滑动窗口算法与互信息量相结合的分割算法(DSMI算法),系统运用GPU的并行计算能力来实现加速检索。实验结果表明,DSMI算法能够加快系统中镜头边界检测的运行效率;同时GPU加速使用户在线检索时间成功达到20倍加速比,满足了项目的实时性要求,具有较高的可扩展性。This video retrieval system focused on how to deal with large-scale data and online real-time retrieval.First,the shot boundary detection algorithm based on mutual information had good performance on retrieval,however,there were such problems in the method as its calculation was complex and inefficient.This paper advanced such a video retrieval method based on the combination of dynamic sliding window and mutual information（DSMI）.Secondly,the system used GPU＇s parallel computing ability to achieve accelerated retrieval.Experiments show that the method can improve the operating efficiency of shot boundary detection,and the system using GPU acceleration to achieve 20 times improvements and highly scalability.

关 键 词：大规模视频检索 动态滑动窗口 互信息量 图形处理器加速 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]