基于事件密集度的交通监控视频存储方法  

Traffic Surveillance Video Storage in HDFS Based on Event Density

作　　者：臧继昆 喻剑[2] 

机构地区：[1]同济大学计算机科学与技术系,上海201804 [2]同济大学嵌入式系统与服务计算教育部重点实验室,上海200092

出　　处：《计算机科学》2015年第5期221-224,229,共5页Computer Science

基　　金：科技部国际合作专项(2012DFG11580)资助

摘　　要：利用HDFS进行大规模交通监控视频的存储和处理是一种可靠、高效、可扩展的数据存储方案。针对HDFS默认的机架感知策略可能造成存储热点这一问题,提出了一种基于事件密集度的交通监控视频放置策略。该策略利用交通视频可按事件类型进行分类这一特征,在数据放置时将数据节点中已存储的各类型的事件视频可能对其造成的负载作为节点的主要评价因素之一,同时结合节点的实时负载、磁盘容量等因素进行综合评价,选择最佳的数据放置节点,从而平衡数据节点的负载。实验表明,基于事件密集度的交通监控视频放置策略可以改善数据节点的吞吐量,提高存储系统性能。Utilizing HDFS to store and process large scale traffic surveillance video is a reliable, highly efficient and scalable solution. However, the default rack awareness data placement strategy in HDFS may cause hotspots when placing data. To address this problem, this paper presented a traffic surveillance video data placement strategy based on traffic event density. The characteristic of traffic surveillance videos allows us to classify them according to traffic event types. When placing new data, the proposed strategy predicts the load of each datanode which is influenced by various of traffic events the datanode stores, then combines the instant load and disk capacity to evaluate each datanode, and choo- ses the most suitable datanode to store new data. Experiments show that the proposed strategy alleviates the hotspot problem and effectively improves the load balancing and throughput in comparison with the default strategy.

关 键 词：交通监控视频 HDFS 交通事件 数据放置 吞吐量 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]